Professional Documents
Culture Documents
Mathematical Inequalities Volume 5 Other Recent Methods For Creating and Solving Inequalities
Mathematical Inequalities Volume 5 Other Recent Methods For Creating and Solving Inequalities
net/publication/329474187
Mathematical Inequalities Volume 5 Other Recent Methods for Creating and Solving
Inequalities
CITATIONS READS
0 512
1 author:
VASILE CIRTOAJE
Petroleum - Gas University of Ploiesti
41 PUBLICATIONS 107 CITATIONS
SEE PROFILE
All content following this page was uploaded by VASILE CIRTOAJE on 28 January 2020.
!"#
!"#$
%
"
!
&
' &
(
)*!
&
"
"*)+,-!
./
2 pqr Method 99
2.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
2.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A Glosar 523
B Index 537
i
ii Vasile Cîrtoaje
Chapter 1
F (a1 , a2 , . . . , an ) : → , ∈ n ,
Proof. Let
A j = (a j1 , a j2 , . . . , a jn ), j = 0, 1, 2, ... ,
where each A j is constructed from the preceding A j−1 by replacing its smallest and
largest elements with their arithmetic mean, and
By hypothesis, we have
1
2 Vasile Cîrtoaje
Lemma. Let
A0 , A1 , A2 , . . .
be an infinite sequence of n-tuples A j = (a j1 , a j2 , . . . , a jn ) ∈ n , where each A j is
constructed from the preceding A j−1 by replacing its smallest and largest elements with
their arithmetic mean. Then
and
a01 + a02 + · · · + a0n
lim A j = (a, a, . . . , a), a= .
j→∞ n
Proof. For any n-tuple A j , define the closed interval j = [a j , b j ], where
| j | = b j − a j .
Clearly, we have
We infer that for any integer j there exists an integer k j such that
n 2
kj ≤ , | j+k j | ≤ | j |.
2 3
Under this assumption, we have
lim | j | = 0,
j→∞
therefore
a01 + a02 + · · · + a0n
lim A j = (a, a, . . . , a), a= .
j→∞ n
To end the proof, let
2a j + b j 2a j + b j a j + 2b j a j + 2b j
j = aj, , j = , , j = , bj .
3 3 3 3
If c1 ∈ j and c2 ∈ j , then
c1 + c2 a j + (a j + 2b j )/3 2a j + b j
> = ,
2 2 3
c1 + c2 (2a j + b j )/3 + b j a j + 2b j
< = ,
2 2 3
therefore
c 1 + c2
∈ j.
2
As a consequence, A j+k j has k j1 − k j elements in j , k j2 − k j elements in j , and the
other elements in j . Since
(k j1 − k j )(k j2 − k j ) = 0,
F (a1 , a2 , . . . , an ) : → , ∈ n ,
From the AM-Theorem and the AM-Corollary, we get the AM1-Theorem and
the AM1-Corollary, respectively.
G(a1 , a2 , . . . , an ) : → , ∈ +n ,
Proof. Let
b1 = a12 , b2 = a22 , . . . , bn = an2
and
F (b1 , b2 , . . . , bn ) = G b1 , b2 , . . . , bn .
Since
a12 + a22 + · · · + an2 = b1 + b2 + · · · + bn ,
G(a1 , a2 , . . . , an ) = G b1 , b2 , . . . , bn = F (b1 , b2 , . . . , bn ),
a2 + a2
a2 + a2
b + b
b + b
1 n 1 n 1 n 1 n
G , a2 , . . . , an−1 , =G , b2 , . . . , bn−1 ,
2 2 2 2
b1 + b n b1 + b n
=F , b2 , . . . , bn−1 , ,
2 n
the AM1-Theorem follows immediately from the AM-Theorem.
G(a1 , a2 , . . . , an ) : → , ∈ +n ,
***
S = {(a1 , a2 , . . . , an ) : a1 + a2 + · · · + an = s, ai ≥ 0, i = 1, 2, . . . , n}.
If
F (a1 , a2 , a3 , . . . , an ) ≥
a + a a + a
1 2 1 2
≥ min F , , a3 , . . . , an , F (0, a1 + a2 , a3 , . . . , an ) (*)
2 2
for all (a1 , a2 , . . . , an ) ∈ S, then
s s
F (a1 , . . . , an−k , an−k+1 , . . . , an ) ≥ min F 0, . . . , 0, , . . . ,
1≤k≤n k k
for all (a1 , a2 , . . . , an ) ∈ S.
Proof The AC-Theorem is clearly true for n = 2. Consider further n ≥ 3. Since the
function F is continuous on the compact set S, F achieves its minimum at one or
more points of the set. We need to show that among these global minimum points
there is one having n − k coordinates equal to zero and k coordinates equal to
s/k, where k ∈ {1, 2, . . . , n}. Using the mathematical induction, it is easy to prove
that this is true if among the global minimum points there is one having either a
coordinate equal to zero or all coordinates equal to s/n. Let
B0 = (b1 , b2 , . . . , bn ), b1 ≤ b2 ≤ · · · ≤ bn ,
0 < b1 < bn .
6 Vasile Cîrtoaje
A1 = (0, b2 , . . . , bn−1 , b1 + bn )
and
b1 + bn b1 + b n
B1 = , b2 , . . . , bn−1 ,
2 2
is also a global minimum point of F over the set S. In the first case, the proof is
completed. In the second case, starting from B1 as a global minimum point of F ,
we repeat the process infinitely to get A∞ with a coordinate equal to zero and, by
Lemma above, s s s
B∞ = , ,..., .
n n n
Since at least one of A∞ and B∞ is a global minimum point of F , the conclusion
follows.
AC-Corollary. Let s > 0 and let F be a symmetric continuous function on the compact
set in n
S = {(a1 , a2 , . . . , an ) : a1 + a2 + · · · + an = s, ai ≥ 0, i = 1, 2, . . . , n}.
If
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an )
for all (a1 , a2 , . . . , an ) ∈ S satisfying
a + a a + a
1 2 1 2
F (a1 , a2 , a3 , . . . , an ) < F , , a3 , . . . , a n , a1 = a2 ,
2 2
then F (a1 , a2 , . . . , an ) is minimum when n − k of the variables a1 , a2 , . . . , an are zero
s
and the other k variables are equal to , where k ∈ {1, 2, . . . , n}.
k
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an ),
AC1-Corollary. Let s > 0 and let G be a symmetric continuous function on the com-
pact set in n
If
G(a1 , a2 , a3 , . . . , an ) ≥ G 0, a12 + a22 , a3 , . . . , an
for all (a1 , a2 , . . . , an ) ∈ S satisfying
a2 + a2
a2 + a2
1 2 1 2
G(a1 , a2 , a3 , . . . , an ) < G , , a3 . . . , an , a1 = a2 ,
2 2
1.2 Applications
5(a3 + b3 + c 3 + d 3 + e3 + f 3 ) + 36 ≥ 11(a2 + b2 + c 2 + d 2 + e2 + f 2 ).
a bc + bcd + cd a + d a b + a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 ≤ 8.
1 1 1 1
+ + + ≤ 1.
5 − a bc 5 − bcd 5 − cd a 5 − d a b
a1 + a2 + · · · + an = n,
1 1 1 n
+ + ··· + ≤ .
k − a1 a2 · · · an−1 k − a2 a3 · · · an k − an a1 · · · an−2 k−1
1 1 1 1 15
+ + + ≤ .
4 − a bc 4 − bcd 4 − cd a 4 − d a b 11
10 Vasile Cîrtoaje
a1 + a2 + · · · + an = n,
a1 + a2 + · · · + an = n.
Then
2 3 n 2
a1 + a23 + · · · + an3 + ai1 ai2 ai3 ≥ a + a22 + · · · + an2 .
3 1≤i <i <i
3 1
1 2 3 ≤n
a1 + a2 + · · · + an = n.
Then
6
a13 + a23 + · · · + an3 + ai1 ai2 ai3 ≥ 2 a12 + a22 + · · · + an2 .
n(n − 3) 1≤i <i <i
1 2 3 ≤n
Arithmetic Mean Method and Arithmetic Compensation Method 11
a1 + a2 + · · · + an = n.
Then
6
(n + 1) a13 + a23 + · · · + an3 + ai1 ai2 ai3 ≥ 2n a12 + a22 + · · · + an2 .
n − 2 1≤i <i <i
1 2 3 ≤n
nm−1 − 1 2
a1 + a22 + · · · + an2 − n ≥ a1m + a2m + · · · + anm − n.
n−1
a3 + b3 + c 3 + d 3 + a bc + bcd + cd a + d a b ≤ 1.
a3 + b3 + c 3 + d 3 + a bc + bcd + cd a + d a b ≥ 2.
28
a3 + b3 + c 3 + d 3 + 3(a + b + c + d) ≤
.
3
a3 + b3 + c 3 + d 3 + 4(a + b + c + d) ≤ 20.
14 Vasile Cîrtoaje
a3 + b3 + c 3 + d 3 + 2 2(a + b + c + d) ≥ 4(2 + 2 ).
a1 + a2 + · · · + an = n,
then
2 n(n − 1) 4
a12 + a22 + · · · + an2 − n2 ≥ a + a24 + · · · + an4 − n .
n2 − n + 1 1
a1 + a2 + · · · + an = m, m ∈ {1, 2, . . . , n},
then
1 1 1 m
+ + ··· + ≥ n− .
1 + a12 1 + a22 1 + an2 2
1 1 1 1 16
+ + + ≥ .
1 + 3a2 1 + 3b2 1 + 3c 2 1 + 3d 2 7
2
1.48. If a1 , a2 , . . . , an are nonnegative real numbers such that a1 + a2 + · · · + an = ,
3
then ai a j 1
≤ .
1≤i≤ j≤n
(1 − a i )(1 − a j ) 4
1.3 Solutions
First Solution. If
F (a, b, c, d) ≥ F (t, b, c, t)
for
a+d
t= ,
2
then, by the AM-Theorem, we have
F (a, b, c, d) ≥ F (1, 1, 1, 1) = 0.
Since
2(b2 + c 2 ) ≥ (b + c)2 ,
it suffices to show that
3(a + c)2
5(a2 + c 2 ) − 6ac + 16d 2 − − 7(a + c)d ≥ 0.
2
Since
5(a2 + c 2 ) − 6ac ≥ (a + c)2 = 4t 2 ,
it suffices to show that
4t 2 + 16d 2 − 6t 2 − 14t d ≥ 0,
Arithmetic Mean Method and Arithmetic Compensation Method 19
(d − t)(8d + t) ≥ 0.
which is equivalent to
Since
3a2 + d 2 − 4 = 12(a − 1)2 , 3a2 + d 2 + 4 = 4(3a2 − 6a + 5),
3a + d − 4 = 12(a − 1)2 (5 − 2a),
3 3
and
13 2 7
16a2 − 26a + 11 = 4a − + > 0.
4 16
5(a3 + b3 + c 3 + d 3 + e3 + f 3 ) + 36 ≥ 11(a2 + b2 + c 2 + d 2 + e2 + f 2 ).
F (a, b, c, d, e, f ) ≥ F (t, b, c, d, t, f )
for
a+e
a ≥ b ≥ c ≥ d ≥ e ≥ f, t= .
2
20 Vasile Cîrtoaje
5(a3 + e3 − 2t 3 ) − 11(a2 + e2 − 2t 2 ) ≥ 0,
45(a + e) − 11(a + b + c + d + e + f ) ≥ 0,
6
a=b=c=d=e= , f =0
5
(or any cyclic permutation).
Second Solution. Apply the AC-Corollary. We will show that
involves
F (a, b, c, d, e, f ) ≥ F (0, 2t, c, d, e, f ),
where
a+b
t= , a = b.
2
Arithmetic Mean Method and Arithmetic Compensation Method 21
(k − 5)(k − 6) ≥ 0.
a bc + bcd + cd a + d a b + a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 ≤ 8.
F (a, b, c, d) ≥ F (t, b, t, d)
for
a+c
a ≥ b ≥ c ≥ d, t= .
2
Taking into account that
a3 + 3a2 d + a6 + 3a4 d 2 ≤ 8,
which is equivalent to
where
a+b
t= , a = b.
2
Write the hypothesis F (a, b, c, d) < F (t, t, c, d) as follows:
bcd + b2 c 2 d 2 ≤ 8.
therefore
64 64 8
bcd + b2 c 2 d 2 ≤ 1+ =8− < 8.
27 27 729
1 1 1 1
+ + + ≤ 1.
5 − a bc 5 − bcd 5 − cd a 5 − d a b
24 Vasile Cîrtoaje
1 1 1 1
F (a, b, c, d) = 1 − − − − ,
5 − a bc 5 − bcd 5 − cd a 5 − d a b
and apply the AC-Corollary. First, we will show that
involves
F (a, b, c, d) ≥ F (0, 2t, c, d),
a+b
where t = , a = b. Write the hypothesis F (a, b, c, d) < F (t, t, c, d) as
2
1 1 1 1 2(5 − cd t) 2
− + − < − .
5 − ct 2 5 − a bc 5 − dt 2 5 − dab (5 − bcd)(5 − cd a) 5 − cd t
c d 2c 2 d 2
+ < .
(5 − a bc)(5 − c t ) (5 − d a b)(5 − d t ) (5 − bcd)(5 − cd a)(5 − cd t)
2 2
Since
c d c d
+ ≥ + ,
(5 − a bc)(5 − c t 2 ) (5 − d a b)(5 − d t 2 ) 5(5 − a bc) 5(5 − d a b)
we get
c d 10c 2 d 2
+ < . (*)
5 − a bc 5 − d a b (5 − bcd)(5 − cd a)(5 − cd t)
Similarly, write the required inequality F (a, b, c, d) ≥ F (0, 2t, c, d) as follows:
1 1 1 1 1 1 1 1
+ − + ≥ − + − ,
5 5 − 2cd t 5 − bcd 5 − cd a 5 − a bc 5 5 − dab 5
10c 2 d 2 c d
≥ + .
(5 − bcd)(5 − cd a)(5 − cd t) 5 − a bc 5 − d a b
we have
1 1 1 1 1 3 27 3
+ + + = + ≤ + < 1.
5 − a bc 5 − bcd 5 − cd a 5 − d a b 5 − bcd 5 71 5
The equality holds for a = b = c = d = 1.
Remark. In the same manner, we can prove the following generalization:
• If a1 , a2 , . . . , an (n ≥ 4) are nonnegative real numbers such that
a1 + a2 + · · · + an = n,
then
1 1 1
+ + ··· + ≤ 1,
n + 1 − a1 a2 · · · an−1 n + 1 − a2 a3 · · · an n + 1 − an a1 · · · an−2
a1 + a2 + · · · + an = n,
1 1 1 n
+ + ··· + ≤ .
k − a1 a2 · · · an−1 k − a2 a3 · · · an k − an a1 · · · an−2 k−1
We assert that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (1)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ) (2)
for any k > en−1 , where
a1 + a2
t= , a1 = a2 .
2
Let us denote
b = a3 a4 · · · an .
Since
1 1 ai
n
−F (a1 , a2 , a3 , . . . , an ) = + + ,
k − a1 b k − a2 b i=3 kai − a1 a2 b
and
2 n
ai
−F (t, t, a3 , . . . , an ) = + ,
k − t b i=3 kai − t 2 b
the hypothesis (1) is equivalent to
2b
n
ai
b(t − a1 a2 )
2
− > 0,
(k − t b)(k − a1 b)(k − a2 b) i=3 (kai − t 2 b)(kai − a1 a2 b)
which involves
2b n
ai
> .
(k − t b)(k − a1 b)(k − a2 b) i=3
(kai − t 2 b)(kai − a1 a2 b)
Because
ai 1
≥ ,
kai − t 2 b k
Arithmetic Mean Method and Arithmetic Compensation Method 27
2k b n
1
> . (3)
(k − t b)(k − a1 b)(k − a2 b) i=3
ka i − a1 a2 b
1 1 n
ai
−F (a1 , a2 , a3 , . . . , an ) = + + ,
k − a1 b k − a2 b i=3 kai − a1 a2 b
and
n−1 1
−F (0, 2t, a3 , . . . , an ) =
+ ,
k k − 2t b
the required inequality (2) can be written as follows:
n
1 1 1 1 1 1 ai 1
− − − − − ≥ − ,
k − 2t b k k − a1 b k k − a2 b k i=3
kai − a1 a2 b k
2t b 2(kt − a1 a2 b)b n
a1 a2 b
− ≥ ,
k − 2t b (k − a1 b)(k − a2 b) i=3
ka i − a1 a2 b
2b(k − t b) n
1
a1 a2 b − ≥ 0.
(k − 2t b)(k − a1 b)(k − a2 b) i=3 kai − a1 a2 b
Because
k − tb k
≥ ,
k − 2t b k − tb
this inequality is true if
2k b n
1
> ,
(k − t b)(k − a1 b)(k − a2 b) i=3
kai − a1 a2 b
kn
en−1 ≤ ,
k+n−1
28 Vasile Cîrtoaje
(n − 1)en−1
k≥ .
n − en−1
(n − 1)en−1
The equality holds for a1 = a2 = · · · = an = 1. In addition, if k = , then
n − en−1
the equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
Remark. For n = 4, we get the following statement:
• If a, b, c, d are nonnegative real numbers such that a + b + c + d = 4 , then
1 1 1 1 4
+ + + ≤ ,
48 − 11a bc 48 − 11bcd 48 − 11cd a 48 − 11d a b 37
4
with equality for a = b = c = d = 1, and also for a = 0 and b = c = d = (or any
3
cyclic permutation).
Solution. The proof is similar to the one of the preceding P 1.5. Finally, it suffices to
prove the original inequality for a = b = c = d = 1, and for a = 0 and b + c + d = 4.
The first case is trivial, while the second case leads to the inequality
1 3 15
+ ≤ ,
4 − bcd 4 11
which is equivalent to
64
bcd ≤ .
27
By the AM-GM inequality, we have
b + c + d 3 64
bcd ≤ = .
3 27
The equality holds for
4
a = 0, b=c=d=
3
Arithmetic Mean Method and Arithmetic Compensation Method 29
a1 + a2 + · · · + an = n.
1 1 1 en−1
+ + ··· + ≤1+ ,
n − a1 a2 · · · an−1 n − a2 a3 · · · an n − an a1 · · · an−2 n(n − en−1 )
n
with equality for a1 = 0 and a2 = · · · = an = (or any cyclic permutation).
n−1
a1 + a2 + · · · + an = n,
1 1 1 n−1 1
+ + ··· + ≤ + .
k − a1 a2 · · · an−1 k − a2 a3 · · · an k − an a1 · · · an−2 k k − en−1
n−1 1
F (a1 , a2 , . . . , an ) + + ≥ 0,
k k − en−1
where
1 1 1
F (a1 , a2 , . . . , an ) = − + + ··· + .
k − a1 a2 · · · an−1 k − a2 a3 · · · an k − an a1 · · · an−2
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an )
for any k > en−1 , where
a1 + a2
t= , a1 = a2 .
2
According to the AC-Corollary, it suffices to prove the original inequality for a1 =
a2 = · · · = an = 1, and for a1 = 0 and a2 + a3 + · · · + an = n. In the first case, the
original inequality becomes
n n−1 1
≤ + ,
k−1 k k − en−1
(n − 1)en−1
k≤ .
n − en−1
1 1
≤ ,
k − a2 a3 · · · an k − en−1
which is equivalent to
a2 a3 · · · an ≤ en−1 .
By the AM-GM inequality, we have
a + a + · · · + a n−1 n n−1
1 2 n−1
a1 a2 · · · an−1 ≤ ≤ = en−1 .
n−1 n−1
The equality holds for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(n − 1)en−1
(or any cyclic permutation). In addition, if k = , then the equality holds
n − en−1
also for a1 = a2 = · · · = an = 1.
1 1 1 1 1 1 2
+ + + + + ≤ .
10 − a b 10 − bc 10 − cd 10 − d a 10 − ac 10 − bd 3
Arithmetic Mean Method and Arithmetic Compensation Method 31
involves
F (a, b, c, d) ≥ F (0, 2t, c, d),
a+b
where t = , a = b. Write the hypothesis F (a, b, c, d) < F (t, t, c, d) as
2
1 1 2 1 1 2
+ − + + − >
10 − ac 10 − bc 10 − t c 10 − ad 10 − bd 10 − t d
1 1
> − .
10 − t 2 10 − a b
Dividing by the positive factor t 2 − a b, the inequality becomes
2c 2 2d 2 1
+ > .
(10 − ac)(10 − bc)(10 − t c) (10 − ad)(10 − bd)(10 − t d) (10 − a b)(10 − t 2 )
Since
1 1
≥ ,
(10 − a b)(10 − t 2 ) 10(10 − a b)
we get
20c 2 20d 2 1
+ > . (*)
(10 − ac)(10 − bc)(10 − t c) (10 − ad)(10 − bd)(10 − t d) 10 − a b
ab 2(10 − t c) 2(10 − t c)
+ −
10(10 − a b) (10 − ac)(10 − bc) 10(10 − 2t c)
2(10 − t d) 2(10 − t d)
+ − ≤ 0,
(10 − ad)(10 − bd) 10(10 − 2t d)
a bc 2 (10 − t c) a bd 2 (10 − t d) ab
+ ≥ .
(10 − ac)(10 − bc)(5 − t c) (10 − ad)(10 − bd)(5 − t d) 10 − a b
This is true if
c 2 (10 − t c) d 2 (10 − t d) 1
+ ≥ .
(10 − ac)(10 − bc)(5 − t c) (10 − ad)(10 − bd)(5 − t d) 10 − a b
Since
10 − t c 20 10 − t d 20
≥ , ≥ ,
5 − tc 10 − t c 5 − td 10 − t d
it suffices to show that
20c 2 20d 2 1
+ ≥ .
(10 − ac)(10 − bc)(10 − t c) (10 − ad)(10 − bd)(10 − t d) 10 − a b
3 27 2
+ ≤ .
10 74 3
We have
2 3 27 11 27 4
− − = − = > 0.
3 10 74 30 74 2220
In the third case, the inequality becomes
1 1
≤ ,
10 − cd 6
cd ≤ 4.
We have 2
c+d
cd ≤ = 4.
2
The equality holds for a = b = c = d = 1, and for a = b = 2 and c = d = 0 (or any
permutation).
Remark. In the same manner, we can prove the following generalizations (Vasile
Cîrtoaje, 2005):
Arithmetic Mean Method and Arithmetic Compensation Method 33
n
−1
m
(a) If k ≥ m , then
n m
−1
m n
G(a1 , a2 , . . . , an ) ≤ G(1, 1, . . . , 1) ;
n
m m −1
m
(b) If < k ≤ m , then
n n m
−1
m n
n n
G(a1 , a2 , . . . , an ) ≤ G( , . . . , , 0 . . . , 0) .
m m
F (a, b, c, d) ≥ 0,
where
5 2 9
F (a, b, c, d) = 3(a3 + b3 + c 3 + d 3 − 4) − a2 + b2 + c 2 + d 2 − + ,
2 4
and apply the AC-Corollary. We will show that
involves
F (a, b, c, d) ≥ F (0, 2t, c, d), (**)
where
a+b
t= , a = b.
2
Using the identities
9(a + b) − 2(4t 2 + a2 + b2 + 2c 2 + 2d 2 − 5) ≤ 0.
Clearly, this inequality follows immediately from (A). As a consequence, (*) implies
(**). Thus, by the AC-Corollary, we have
where
16(k − 1)(4 − k)
f (k) =
k2
is the value of F for the case where 4 − k of the variables a, b, c, d are zero and the
other k variables are equal to 4/k. We have
32
f (1) = f (4) = 0, f (2) = 8, f (3) = ,
9
therefore F (a, b, c, d) ≥ 0.
The equality holds for a = b = c = 0 and d = 4 (or any cyclic permutation), and
also for a = b = c = d = 1.
F (a, b, c, d) ≥ F (t, b, t, d)
for
a+c
a ≥ b ≥ c ≥ d, t= .
2
Using the identities
Since
4t 2 +2a2 +2c 2 +4b2 +4d 2 = (a + c)2 +2(a2 + c 2 )+4(b2 + d 2 ) ≤ 3(a + c)2 +4(b + d)2 ,
36 Vasile Cîrtoaje
Substituting
x = a + c, y = b + d,
we need to prove that
21x − 3x 2 − 4 y 2 − 14 ≥ 0
for x + y = 4, x ≥ y. We have
which is equivalent to
involves
F (a, b, c, d) ≥ F (0, 2t, c, d), (**)
where
a+b
t= , a = b.
2
As shown at the first solution, (*) involves
Since
21(a + b) − 2(4t 2 + a2 + b2 + 2c 2 + 2d 2 + 7) ≤ 0.
Clearly, this inequality follows immediately from (A). As a consequence, (*) implies
(**). Thus, by the AC-Corollary, we have
where
16(k − 3)(k − 4)
f (k) =
k2
is the value of F for the case where 4 − k of the variables a, b, c, d are zero and the
other k variables are equal to 4/k. We have
therefore F (a, b, c, d) ≥ 0.
a1 + a2 + · · · + an = n.
Then
2 3 n 2
a1 + a23 + · · · + an3 + ai1 ai2 ai3 ≥ a + a22 + · · · + an2 .
3 1≤i <i <i
3 1
1 2 3 ≤n
Solution. Let
b = a3 + · · · + a n
and
F (a1 , a2 , . . . , an ) = k1 a13 + · · · + an3 + ai1 ai2 ai3 − k2 n(a12 + · · · + an2 ),
1≤i1 <i2 <i3 ≤n
38 Vasile Cîrtoaje
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an )
where
a1 + a2
t= , a1 = a2 .
2
Since
ai1 ai2 ai3 = a1 a2 b + (a1 + a2 ) a i1 a i2 + a i1 a i2 a i3 ,
1≤i1 <i2 <i3 ≤n 3≤i1 <i2 ≤n 3≤i1 <i2 <i3 ≤n
f (k) ≥ 0,
and
a13 + a23 + · · · + an3 + 3 a i1 a i2 a i3 ≥ ai1 ai2 (ai1 + ai2 ).
1≤i1 <i2 <i3 ≤n 1≤i1 <i2 ≤n
with equality for a1 = a2 and a3 = 0 (or any cyclic permutation), and also for
a1 = a 2 = a 3 .
a1 + a2 + · · · + an = n.
Then
6
a13 + a23 + · · · + an3 + ai1 ai2 ai3 ≥ 2 a12 + a22 + · · · + an2 .
n(n − 3) 1≤i <i <i
1 2 3 ≤n
Solution. As shown at the preceding P 1.11, it suffices to show that f (k) ≥ 0 for
all k ∈ {1, 2, . . . , n}, where
3(2k2 + 1) 2(3k1 + 1)
f (k) = 1 − + ,
k k2
with
n(n − 3) n−3
k1 = , k2 = .
6 3
We get
2n − 3 (n − 1)(n − 2) (k − n + 1)(k − n + 2)
f (k) = 1 − + = .
k k2 k2
Clearly, for k ∈ {1, 2, . . . , n}, we have
f (k) ≥ 0,
a1 + a2 + · · · + an = n.
Then
6
(n + 1) a13 + a23 + · · · + an3 + ai1 ai2 ai3 ≥ 2n a12 + a22 + · · · + an2 .
n − 2 1≤i <i <i
1 2 3 ≤n
Solution. As shown at P 1.11, it suffices to show that f (k) ≥ 0 for all k ∈ {1, 2, . . . , n},
where
3(2k2 + 1) 2(3k1 + 1)
f (k) = 1 − + ,
k k2
with
(n + 1)(n − 2) n−2
k1 = , k2 = .
6 3
Arithmetic Mean Method and Arithmetic Compensation Method 41
We get
2n − 1 n(n − 1) (k − n)(k − n + 1)
f (k) = 1 − + = .
k k2 k2
Clearly, for k ∈ {1, 2, . . . , n}, we have
f (k) ≥ 0,
a1 = a2 = · · · = an = 1.
2 2
a3 + b3 + c 3 + d 3 + a bc + bcd + cd a + d a b ≥ a (b + c + d),
3
where a, b, c, d are nonnegative real numbers. The equality holds for a = b = c
and d = 0 (or any cyclic permutation), and also for a = b = c = d.
Assume that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
42 Vasile Cîrtoaje
where
(k − m)(k − m + 1)
f (k) =
k2
is the value of F for a1 = · · · = an−k = 0 and an−k+1 = · · · = an = 1/k. Obviously,
for k ∈ {1, 2, . . . , n}, we have
f (k) ≥ 0,
with equality when k = m − 1 (m ≥ 2) and k = m (m ≤ n). From f (k) ≥ 0 for
k ∈ {1, 2, . . . , n}, it follows that F (a1 , a2 , . . . , an ) ≥ 0.
To prove that (*) implies (**), we write (*) as
Remark. Actually, the inequality holds for all m ∈ [0, 1] ∪ {2, 3, . . . , n} ∪ [n+ 1, ∞),
but it does not hold for m ∈ (1, 2) ∪ (2, 3) ∪ · · · ∪ (n, n + 1).
for
a1 + an−1
a1 ≤ a2 ≤ · · · ≤ an−1 ≤ an , t= .
2
Write this inequality as follows:
(n + 1)(a12 + an−1
2
− 2t 2 ) − (a13 + an−1
3
− 2t 3 ) ≥ 0,
We have
a1 = · · · = an−1 = 0, an = n
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ), (**)
44 Vasile Cîrtoaje
where
a1 + a2
t= , a1 = a2 .
2
Then, by AC-Corollary, we have
where
n2 n3 n2 (k − 1)(n − k)
f (k) = (n + 1) − n2 − 2 =
k k k2
is the value of F for a1 = · · · = an−k = 0 and an−k+1 = · · · = an = n/k. Obviously,
f (k) ≥ 0
for
a1 + an−1
a1 ≤ a2 ≤ · · · ≤ an−1 ≤ an , t= .
2
Arithmetic Mean Method and Arithmetic Compensation Method 45
Since
a12 + an−1
2
− 2t 2 = 2(t 2 − a1 an−1 ),
1
a14 + an−1
4
− 2t 4 = (t 2 − a1 an−1 )(7a12 + 7an−1 2
+ 10a1 an−1 ),
2
the inequality F (a1 , a2 , . . . , an−2 , an−1 , an ) ≥ F (t, a2 , . . . , an−2 , t, an ) can be written
as follows:
Since
Since
(n − 1)a12 + an2 − n = n(n − 1)(a1 − 1)2 ,
(n−1)a14 +an4 −n = n(n−1)(a1 −1)2 (n2 − 3n + 3)a12 − 2(n2 − n − 1)a1 + n2 + n + 1 ,
the inequality is equivalent to
It is true because
(n2 − 3n + 3)n n3 − n2 − 3n + 2
2(n2 −n−1)−(n2 −3n+3)a1 ≥ 2(n2 −n−1)− = > 0.
n−1 n−1
The equality holds when a1 = a2 = · · · = an = 1, and also when
a1 = · · · = an−1 = 0, an = n
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
Then, by AC-Corollary, we have
where
4
n2 n n2 (k − 1)(n − k)(nk + n + k)
f (k) = (n2 + n + 1) −n − 3 −n =
k k k3
is the value of F for a1 = · · · = an−k = 0 and an−k+1 = · · · = an = n/k. Obviously,
f (k) ≥ 0
a12 + a22 − (a1 + a2 )2 = −2a1 a2 , a14 + a24 − (a1 + a2 )4 = −a1 a2 (4a12 + 4a22 + 6a1 a2 ),
Arithmetic Mean Method and Arithmetic Compensation Method 47
nm−1 − 1 2
a1 + a22 + · · · + an2 − n ≥ a1m + a2m + · · · + anm − n.
n−1
(Vasile C., 2002)
Solution. Let
nm−1 − 1
r= .
n−1
We need to prove that F (a1 , a2 , . . . , an ) ≥ 0, where
F (a1 , a2 , . . . , an ) = r a12 + a22 + · · · + an2 − n − a1m + a2m + · · · + anm − n .
Assume that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
Then, by AC-Corollary, we have
where
n2 nm
f (k) = r − n − m−1 + n
k k
48 Vasile Cîrtoaje
a1m + a2m − 2t m
> 2r. (A)
t 2 − a1 a2
For the nontrivial case a1 a2 = 0, the desired inequality (**) is equivalent to
a1m + a2m − (a1 + a2 )m
≤ r,
a12 + a22 − (a1 + a2 )2
a1 = 1 + x, a2 = 1 − x, 0 ≤ x < 1,
We have
f (x) = m (1 + x)m−1 − (1 − x)m−1 − (2m+1 − 4)x,
f (t) = m(m − 1) (1 + x)m−2 + (1 − x)m−2 − 2m+1 + 4,
f (x) = m(m − 1)(m − 2) (1 + x)m−3 − (1 − x)m−3 .
Since f > 0, f is strictly increasing. Because
there exists x 1 ∈ (0, 1) such that f (x 1 ) = 0, f (x) < 0 for x ∈ [0, x 1 ) and
f (x) > 0 for x ∈ (x 1 , 1]. Consequently, f’ is strictly decreasing on [0, x 1 ] and
strictly increasing on [x 1 , 1]. Since
there exists x 2 ∈ (x 1 , 1) such that f (x 2 ) = 0, f (x) < 0 for x ∈ (0, x 2 ) and f (x) >
0 for x ∈ (x 2 , 1]. Thus, f is strictly decreasing on [0, x 2 ] and strictly increasing on
[x 2 , 1]. Since f (0) = f (1) = 0, it follows that f (x) ≤ 0 for 0 ≤ x ≤ 1.
The equality holds when a1 = a2 = · · · = an = 1, and also when
a1 = · · · = an−1 = 0, an = n
5
F (a, b, c, d, e) = a2 + b2 + c 2 + d 2 + e2 − 5 − (1 − a bcd e).
4
So, we need to show that
F (a, b, c, d, e) ≥ F (t, b, c, t, e)
for
a+d
a ≥ b ≥ c ≥ d ≥ e, t= .
2
We have
5
F (a, b, c, d, e) − F (t, b, c, t, e) = a2 + d 2 − 2t 2 − bce(t 2 − ad)
4
5
= 2(t 2 − ad) − bce(t 2 − ad)
4
1
= (t 2 − ad)(8 − 5bce).
4
It suffices to show that
8 − 5bce ≥ 0,
50 Vasile Cîrtoaje
hence
8(a + b + c + d + e)3 ≥ 8 · 27 · (4bce) = 864bce ≥ 625bce.
According to AM-Corollary, it suffices to prove the original inequality for a = b =
c = d. Thus, we need to show that 4a + e = 5 implies
5
4a2 + e2 − 5 ≥ (1 − a4 e),
4
which is equivalent to
5
20(1 − a)2 ≥ (1 − a)2 (4a3 + 3a2 + 2a + 1),
4
(1 − a)2 (15 − 2a − 3a2 − 4a3 ) ≥ 0,
(1 − a)2 (3 + 2a + a2 )(5 − 4a) ≥ 0,
(1 − a)2 (3 + 2a + a2 )e ≥ 0.
5
The equality holds for a = b = c = d = e = 1, and also for a = b = c = d =
4
and e = 0 (or any cyclic permutation).
where
f (n) = F (1, 1, . . . , 1) = 0
and
n2 n2
f (k) = − , k ∈ {1, 2, . . . , n − 1},
k n−1
is the value of F for a1 = · · · = an−k = 0 and an−k+1 = · · · = an = n/k. Since
f (k) ≥ 0 for k ∈ {1, 2, . . . , n − 1} (with equality when k = n − 1), we have
F (a1 , a2 , . . . , an ) ≥ 0.
45
F (a, b, c, d, e) = a3 + b3 + c 3 + d 3 + e3 − 5 − (1 − a bcd e).
16
So, we need to show that
F (a, b, c, d, e) ≥ F (t, b, c, t, e)
52 Vasile Cîrtoaje
for
a+d
a ≥ b ≥ c ≥ d ≥ e, t= .
2
We have
45
F (a, b, c, d, e) − F (t, b, c, t, e) = a3 + d 3 − 2t 3 −bce(t 2 − ad)
16
45
= 3(t 2 − ad)(a + d) − bce(t 2 − ad)
16
3 2
= (t − ad)(16a + 16d − 15bce).
16
It suffices to show that
a + d ≥ bce,
which is equivalent to the homogeneous inequality
Let
x= be.
Since
a + d ≥ b + e ≥ 2x, a + b + c + d + e ≥ 2b + c + 2e ≥ 4x + c,
where
f (n) = F (1, 1, . . . , 1) = 0
and
n3 n3
f (k) = − , k ∈ {1, 2, . . . , n − 1},
k 2 (n − 1)2
is the value of F for a1 = · · · = an−k = 0 and an−k+1 = · · · = an = n/k. Since
f (k) ≥ 0 for k ∈ {1, 2, . . . , n − 1} (with equality when k = n − 1), we have
F (a1 , a2 , . . . , an ) ≥ 0.
1 3
F (a1 , a2 , . . . , an ) = a + a23 + · · · + an3 − n − a12 + a22 + · · · + an2 − n
n 1
n−2 2
− (a1 a2 · · · an − 1).
2
If
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ), (**)
where
a1 + a2
, t= a1 = a2 ,
2
then, by the AC-Corollary, we have
where
f (n) = F (1, 1, . . . , 1) = 0
and
2
1 n3 n n − 2 2 (k − 2)2 n2
f (k) = − n − − n + = , k ∈ {1, 2, . . . , n − 1},
n k2 k 2 4k2
F (a1 , a2 , . . . , an ) ≥ 0.
respectively. Clearly, (A) implies (B), therefore (*) implies (**). This completes the
proof.
The equality holds when a1 = a2 = · · · = an = 1, and also when
n(n − 1) 3
F (a1 , a2 , . . . , an ) = a1 + a23 + · · · + an3 − n − (a12 + a22 + · · · + an2 − n)2 .
n+1
If
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, 2t, a3 , . . . , an ), (**)
where
a1 + a2
, t= a1 = a2 ,
2
then, by the AC-Corollary, we have
F (a1 , a2 , . . . , an ) ≥ 0.
a1 = · · · = an−1 = 0, an = n
with
2k32 ≥ k1 > 0, 4k3 ≥ k4 k3 > 0.
We will show that
F (a, b, c, d) < F (t, t, c, d) (*)
involves
F (a, b, c, d) ≥ F (0, 2t, c, d), (**)
where
a+b
t= , a = b.
2
Using the identities
1
a2 + b2 − 2t 2 = 2(t 2 − a b), a4 + b4 − 2t 4 = (t 2 − a b)(7a2 + 7b2 + 10a b),
2
we may write (*) as
a2 + b2
k1 (a4 + b4 − 2t 4 ) − 2k3 (a2 + b2 − 2t 2 ) k3 (t 2 + + c 2 + d 2 ) − k4 < 0,
2
k1 (7a + 7b + 10a b)
2 2
a + b2
2
(t 2 − a b) − 4k3 k3 (t 2 + + c 2 + d 2 ) − k4 < 0,
2 2
k1 (7a2 + 7b2 + 10a b) a 2
+ b 2
− 4k3 k3 (t 2 + + c 2 + d 2 ) − k4 < 0. (A)
2 2
Since
which is true because k1 ≤ 2k32 . As a consequence, (*) implies (**). Thus, by the
AC-Corollary, we have
F (a, b, c, d) ≥ min f (k),
1≤k≤n
where
2 2
4 4 44 k1 4 k3
f (k) = min F 0, . . . , 0, , . . . , = 3 + k2 − − k4
1≤k≤4 k k k k
is the value of F for the case where 4 − k of the variables a, b, c, d are zero and the
other k variables are equal to 4/k. Since
42 k3
− k4 ≥ 4k3 − k4 ≥ 0,
k
we only need to show that g(k) ≥ 0 for k ∈ {1, 2, 3, 4}, where
256k 16k3
1
g(k) = + k2 − + k4 .
k3 k
For the original inequality, we have
313 5
k1 = 2, k2 = , k3 = 1, k4 = ,
81 9
5
2k32 − k1 = 0, 4k3 − k4 = 4 − >0, k3 > 0,
9
512 313 16 5
g(k) = + − + , k ∈ {1, 2, 3, 4}.
k3 81 k 9
Since
313 5
with
4 16
k1 = , k2 = − , k3 = 1, k4 = 4,
7 7
4
2k32 − k1 = 2 − > 0, 4k3 − k4 = 0 k3 > 0.
7
As shown in the proof of the preceding P 1.24, we only need to show that g(k) ≥ 0
for k = 1, 2, 3, 4, where
256k 16k3 4
64 − k3 16
1
g(k) = + k2 − + k4 = − + 4.
k3 k k 7k k
We have
63
g(1) = 4 − 12 = 0,
7
56
g(2) = 2 − 4 = 0,
14
4
37 16 4 16
g(3) = − +4> − + 4 = 0,
3 21 3 3 3
g(4) = 0.
The equality holds for a = b = c = 0 and d = 4 (or any cyclic permutation), for
a = b = 0 and c = d = 2 (or any permutation), and also for a = b = c = d = 1.
with
4 3 16
k1 = 1, k2 = 0, k3 =
− , k4 =
− 8,
3 2 3
91 − 48 3
2k3 − k1 =
2
> 4(3 − 2 3 ) > 0, 4k3 − k4 = 2 > 0.
6
As shown in the proof of P 1.24, we only need to show that g(k) ≥ 0 for k =
1, 2, 3, 4, where
256k 16k3
1
g(k) = + k2 − + k4 = 8h(k),
k3 k
2 2 4 3
h(k) =
−
− 1 + − 1.
k k 3 k k
We have
with
1
k1 = 1, k2 = 0, k3 = 2 −
, k4 = 8(2 − 2 ),
2
As shown in the proof of P 1.24, we only need to show that g(k) ≥ 0 for k =
1, 2, 3, 4, where
256k 16k3
1
g(k) = + k2 − + k4 = 16h(k),
k3 k
1 2 1 1
h(k) =
+ 1 − −
1 − .
k k k 2 k
We have
h(1) = 0,
h(2) = 0,
1 1
h(3) = 1 +
− 2 > 0,
3 3
5−3 2
h(4) = > 0.
8
The equality holds when a = b = c = 0 and d = 4 (or any cyclic permutation),
and also when a = b = 0 and c = d = 2 (or any permutation).
with
3 2 1 1
k1 = 1, k2 = 0, k3 =
−
, k4 = 16
−
,
2 3 2 3
2 1
2k32 − k1 > 0, 4k3 − k4 = 4
−
> 0.
3 2
As shown in the proof of P 1.24, we only need to show that g(k) ≥ 0 for k =
1, 2, 3, 4, where
256k 16k3
1
g(k) = + k2 − + k4 = 16h(k),
k3 k
1 1 3 1 2
h(k) =
+
1 − −
1− .
k k 2 k 3 k
62 Vasile Cîrtoaje
We have
1 3 1
h(1) = 1 − 2 +
> 1 − + = 0,
3 2 2
h(2) = 0,
h(3) = 0,
1 1 1 1
h(4) = +
−
> 0.
2 4 2 2 3
The equality holds when a = b = 0 and c = d = 2 (or any permutation), and
also when a = 0 and b = c = d = 4/3 (or any cyclic permutation).
a3 + b3 + c 3 + d 3 + a bc + bcd + cd a + d a b ≤ 8.
G(a, b, c, d) ≥ G(t, b, t, d)
for
a2 + c 2
a ≤ b ≤ c ≤ d, t= .
2
Write this inequality as follows:
4t 2 − (a + c)2 (a3 + c 3 )2 − 4t 6
(t 2 − ac)(b + d) + bd − 3 ≥ 0,
2t + a + c a + c 3 + 2t 3
(a − c)2 (b + d) (a − c)2 (a − c)2 [(a2 + ac + c 2 )2 − 3a2 c 2 ]
+ bd − ≥ 0,
2 2t + a + c 2(a3 + c 3 + 2t 3 )
which is true if
2bd (a2 + ac + c 2 )2
b+d+ ≥ 3 .
2t + a + c a + c 3 + 2t 3
Since
b + d ≥ a + c, bd ≥ ac,
2t + a + c ≤ 3(a + c)
Arithmetic Mean Method and Arithmetic Compensation Method 63
and
1
a3 + c 3 + 2t 3 ≥ a3 + c 3 + (a + c)(a2 + c 2 ) ≥ a3 + c 3 + ac(a + c) = (a + c)(a2 + c 2 ),
2
it suffices to show that
2ac (a2 + ac + c 2 )2
a+c+ ≥ ,
3(a + c) (a + c)(a2 + c 2 )
which is equivalent to
3(a2 + ac + c 2 )2
3(a2 + c 2 ) + 8ac ≥ ,
a2 + c 2
2ac(a2 + c 2 ) ≥ 3a2 c 2 ,
2ac(a − c)2 + a2 c 2 ≥ 0.
By the AM1-Corollary, it suffices to prove the original inequality for a = b = c;
that is, to show that 3a2 + d 2 = 4 involves
4a3 + d 3 + 3a2 d ≤ 8.
4a3 + 4d ≤ 8,
2 − a3 ≥ d.
Since 3a2 ≤ 4, we have
8
2 − a3 ≥ 2 −
> 0.
3 3
Thus, we only need to show that
(2 − a3 )2 ≥ d 2 ,
which is equivalent to
(2 − a3 )2 ≥ 4 − 3a2 ,
a2 (a4 − 4a + 3) ≥ 0,
a2 (a − 1)2 (a2 + 2a + 3) ≥ 0.
The equality holds for a = b = c = d = 1, and also for a = b = c = 0 and d = 2 (or
any cyclic permutation).
64 Vasile Cîrtoaje
a3 + b3 + c 3 + d 3 + a bc + bcd + cd a + d a b ≥ 2.
involves
G(a, b, c, d) ≥ G(0, 2 t, c, d)
for any k > 0 and real m, where
a2 + b2
t= , a = b.
2
Write the hypothesis G(a, b, c, d) < G(t, t, c, d) as
(a3 + b3 )2 − 4t 6 4t 2 − (a + b)2
− k(t 2 − a b)(c + d) − k · cd < 0,
a + b + 2t
3 3 3 2t + a + b
(a − b) [a + b + 2a b(a + b )] k(a − b) (c + d)
2 4 4 2 2 2
k(a − b)2
− − cd < 0,
2(a + b + 2t )
3 3 3 2 2t + a + b
a4 + b4 + 2a b(a2 + b2 ) 2cd
<k c+d+ . (*)
a + b + 2t
3 3 3 2t + a + b
(a + b)2 − 2t 2 8t 6 − (a3 + b3 )2
ka b(c + d) + k ·
cd ≥
,
a+b+ 2 t 2 2 t 3 + a3 + b3
2ka bcd a2 b2 (3a2 + 3b2 − 2a b)
ka b(c + d) +
≥
,
a+b+ 2 t 2 2 t 3 + a3 + b3
2kcd a b(3a2 + 3b2 − 2a b)
k(c + d) +
≥
.
a+b+ 2t 2 2 t 3 + a3 + b3
This inequality is true if
which is equivalent to
(a2 + b2 )(a2 + b2 − a b) ≥ 0.
By the AC1-Corollary, it suffices to prove that
2 2 2 2
2
2
2
G , , , ≥ 0, G 0, , , ≥ 0,
2 2 2 2 3 3 3
G(0, 0, 1, 1) ≥ 0, G(0, 0, 0, 2) ≥ 0.
We have
2 2 2 2
2
2
2 8 6
G , , , = 4 2 − 2, G 0, , , = − 2,
2 2 2 2 3 3 3 9
G(0, 0, 1, 1) = 0, G(0, 0, 0, 2) = 2 2 − 2.
The equality holds for a = b = 0 and c = d = 1 (or any permutation).
then
3
3
G 0, 0, , ≥ 0, G(0, 0, 0, 3) ≥ 0.
2 2
We have
3 3 3 3 15 3
G , , , = − 11 > 0, G (0, 1, 1, 1) = 0,
2 2 2 2 2
3
3
3
G(a, b, c, d) ≥ G(t, b, t, d)
for
a2 + c 2
a ≥ b ≥ c ≥ d, t= .
2
Write this inequality as follows:
4[(a3 + c 3 )2 − 4t 6 ] 4t 2 − (a + c)2
− (t 2 − ac)(b + d) − bd ≥ 0
a + c + 2t
3 3 3 2t + a + c
4(a − c)2 [a4 + b4 + 2a b(a2 + b2 )] (a − c)2 (b + d) (a − c)2
− − bd ≥ 0,
2(a + c + 2t )
3 3 3 2 2t + a + c
which is true if
4a4 + 4c 4 + 8ac(a2 + c 2 ) 2bd
≥ b+d+ .
a3 + c 3 + 2t 3 2t + a + c
Since
2t 3 ≤ a3 + c 3 , 2t ≥ a + c,
Arithmetic Mean Method and Arithmetic Compensation Method 67
which is equivalent to
2a4 + 2c 4 + 4ac(a2 + c 2 )
≥ (a + c)(b + d) + bd,
a2 + c 2 − ac
Since b + d ≤ a + c and bd ≤ ac, we only need to prove that
2a4 + 2c 4 + 4ac(a2 + c 2 )
≥ (a + c)2 + ac,
a2 + c 2 − ac
which is equivalent to
a4 + c 4 + 2ac(a2 + c 2 ) + a2 c 2 ≥ 0.
By the AM1-Corollary, it suffices to prove the original inequality for a = b = c;
that is, to show that 3a2 + d 2 = 4 involves
which is equivalent to
13a3 + 4d 3 + 3a2 d ≥ 20,
13a3 + 4d(4 − 3a2 ) + 3a2 d ≥ 20,
(16 − 9a2 )d ≥ 20 − 13a3 . (*)
Since 3a ≤ 4 involves 16 − 9a > 0, we only need to show that
2 2
20
for a3 ≤ , which is equivalent to
13
312
a(1 − a)2 f (a) ≥ 0, f (a) =+ 624 − 24a − 412a2 − 206a3 .
a
20
Since f is decreasing, it suffices to show that f (a) ≥ 0 for a3 = . This is true
13
20
because (*) holds for a =
3
.
13
The equality holds for a = b = c = d = 1.
68 Vasile Cîrtoaje
then
G(0, 0, 2, 2) ≥ 0, G(0, 0, 0, 2) ≥ 0.
We have
2 2 2 104
G(1, 1, 1, 1) = 0,G 0,
,
,
=
− 20 > 0,
3 3 3 3 3
G(a, b, c, d) ≥ G(t, b, t, d)
for k ≥ 3 and
a2 + c 2
a ≤ b ≤ c ≤ d, t= .
2
Write this inequality as follows:
k(2t − a − c) − (a3 + c 3 − 2t 3 ) ≥ 0,
Arithmetic Mean Method and Arithmetic Compensation Method 69
k(a2 + b2 + c 2 + d 2 ) a4 + c 4 + 2ac(a2 + c 2 )
≥ 3 ,
2(2t + a + c) a + c 3 + (a2 + c 2 )t
Since
k(a2 + b2 + c 2 + d 2 ) ≥ 2k(a2 + c 2 ) ≥ 6(a2 + c 2 ),
it suffices to show that
3(a2 + c 2 ) a4 + c 4 + 2ac(a2 + c 2 )
≥ 3 ,
2t + a + c a + c 3 + (a2 + c 2 )t
which is equivalent to
x 3 + (4 − x 2 ) y + 9x + 3 y ≤ 28,
(7 − x 2 ) y ≤ 28 − 9x − x 3 .
Since x ≤ 2 implies
28 − 9x − x 3 = 2 + 9(2 − x) + (8 − x 3 ) > 0,
70 Vasile Cîrtoaje
(7 − x 2 )2 y 2 ≤ (28 − 9x − x 3 )2 ,
which is equivalent to
(7 − x 2 )2 (12 − 3x 2 ) ≤ (28 − 9x − x 3 )2 ,
f (x) = x 4 + 2x 3 − 6x 2 − 28x + 49
= (x − 2)2 (x 2 + 6x + 14) + 4x − 7
= (2 − x)(24 − 2x − 4x 2 − x 3 ) + 1.
a3 + b3 + c 3 + d 3 + 4(a + b + c + d) ≤ 20.
G(a, b, c, d) ≥ G(t, b, t, d)
for k ≥ 3 and
a2 + c 2
a ≤ b ≤ c ≤ d, t= .
2
Arithmetic Mean Method and Arithmetic Compensation Method 71
which is equivalent to
which is equivalent to
a3 + b3 + c 3 + d 3 + 2 2 (a + b + c + d) ≥ 4(2 + 2 ).
involves
G(a, b, c, d) ≥ G(0, 2 t, c, d)
for any k > 0 and real m, where
a2 + b2
t= , a = b.
2
(a3 + b3 )2 − 4t 6 4t 2 − (a + b)2
−k· cd < 0,
a + b + 2t
3 3 3 2t + a + b
(a − b)2 [a4 + b4 + 2a b(a2 + b2 )] k(a − b)2
− < 0,
2(a + b + 2t )
3 3 3 2t + a + b
[a4 + b4 + 2a b(a2 + b2 )](2t + a + b)
< 2k. (*)
a3 + b3 + 2t 3
k(a + b − 2 t) ≥ 2 2 t 3 − a3 − b3 ,
and
3(2t + a + b) 2(a + b + 2 t)
≥
.
a3 + b3 + 2t 3 2 2 t 3 + a3 + b3
The first inequality is equivalent to
G(0, 0, 2, 2) ≥ 0, G(0, 0, 0, 2) ≥ 0.
We have
G(1, 1, 1, 1) = 4( 2 − 1) > 0,
2 2 2
2
G 0,
,
,
= 4( 3 − 1) 2−
> 0,
3 3 3 3
G(0, 0, 2, 2) = 0, G(0, 0, 0, 2) = 0.
The equality holds for a = b = 0 and c = d = 2 (or any permutation), and also
for a = b = c = 0 and d = 2 (or any cyclic permutation).
then
G(a, b, c, d) ≥ G(0, 2 t, c, d)
for any k > 0 and real m, where
a2 + b2
t= , a = b.
2
74 Vasile Cîrtoaje
G(0, 0, 2, 2) ≥ 0, G(0, 0, 0, 2) ≥ 0.
We have
2 2 2 2
G(1, 1, 1, 1) = 4( 2 − 1)
− 1 > 0, G 0,
,
,
=0
3 3 3 3
1
G(0, 0, 2, 2) = 0, G(0, 0, 0, 2) = 4(2 − 2) 1 −
> 0.
3
2
The equality holds for a = 0 and b = c = d =
(or any cyclic permutation),
3
and also for a = b = 0 and c = d = 2 (or any permutation).
G(a, b, c, d) ≥ G(t, b, t, d)
4
for k ≤ and
3
a2 + c 2
a ≥ b ≥ c ≥ d, t= .
2
Write this inequality as follows:
(a3 + c 3 − 2t 3 ) − k(2t − a − c) ≥ 0,
which is true if
a4 + c 4 + 2ac(a2 + c 2 ) 2k
≥ .
a3 + c 3 + (a2 + c 2 )t 2t + a + c
Write this inequality in the homogeneous form
a4 + c 4 + 2ac(a2 + c 2 ) k(a2 + b2 + c 2 + d 2 )
≥ ,
a3 + c 3 + (a2 + c 2 )t 2(2t + a + c)
Since
8(a2 + c 2 )
k(a2 + b2 + c 2 + d 2 ) ≤ 2k(a2 + c 2 ) ≤ ,
3
it suffices to show that
a4 + c 4 + 2ac(a2 + c 2 ) 4(a2 + c 2 )
≥ . (*)
a3 + c 3 + (a2 + c 2 )t 3(2t + a + c)
A = a4 + c 4 + 6ac(a2 + c 2 ) − 4a2 c 2 .
a+c a+c
Since A > 0 and t ≥ , it suffices to prove the inequality (*) for t = . That
2 2
means to show that
a4 + c 4 + 2ac(a2 + c 2 ) 4(a2 + c 2 )
≥ ,
a3 + c + (a + c )(a + c)/2
3 2 2 6(a + c)
which is true if
a4 + c 4 + 2ac(a2 + c 2 ) 2(a2 + c 2 )
≥ ,
a2 + c − ac + (a + c )/2
2 2 2 3
which is equivalent to
a4 + c 4 + 2ac(a2 + c 2 ) a2 + c 2
≥ ,
3(a2 + c 2 ) − 2ac 3
ac(4a2 + 4c 2 − 3ac) ≥ 0.
According to the AM1-Corollary, it suffices to prove the original inequality for a =
b = c. Write the original inequality
2
a3 + b3 + c 3 + d 3 + k(a + b + c + d) ≥ 4(1 + k), k=
,
3
in the homogeneous inequality
4(a3 + b3 + c 3 + d 3 )
+ k(a + b + c + d) ≥ 2(1 + k) a2 + b2 + c 2 + d 2 .
a2 + b2 + c 2 + d 2
76 Vasile Cîrtoaje
G(a, b, c, d) ≥ G(0, 2 t, c, d)
for any k > 0, where
a2 + b2
t= , a = b.
2
Therefore, by the AC1-Corollary, it suffices to prove that
2 2 2
G(1, 1, 1, 1) ≥ 0, G 0,
,
,
≥ 0,
3 3 3
G(0, 0, 2, 2) ≥ 0, G(0, 0, 0, 2) ≥ 0.
We have
2 2 2
G(1, 1, 1, 1) = 0,
G 0,
,
,
=0
3 3 3
1 1 1
G(0, 0, 2, 2 ) = 4(2 − 2 )
−
> 0, G(0, 0, 0, 2) = 4 1 −
> 0.
2 3 3
Arithmetic Mean Method and Arithmetic Compensation Method 77
a1 + a2 + · · · + an = n,
then
2 n(n − 1) 4
a12 + a22 + · · · + an2 − n2 ≥ a + a24 + · · · + an4 − n .
n2 − n + 1 1
(Vasile C., 2006)
with
n(n − 1)
. k=
n2 − n + 1
Without loss of generality, assume that
a1 ≤ a2 ≤ · · · ≤ an
or
a1 ≥ a2 ≥ · · · ≥ an ,
such that
an2 = max{a12 , a22 , . . . , an2 }.
If
a1 + an−1
F (a1 , a2 , . . . , an−2 , an−1 , an ) ≥ F (t, a2 , . . . , an−2 , t, an ) , t= ,
2
then, by the AM-Corollary, it suffices to prove the original inequality for
a1 = a2 = · · · = an−1 .
and
(n−1)a4 +an4 −n = n(n−1)(a−1)2 (n2 − 3n + 3)a2 − 2(n2 − n − 1)a + n2 + n + 1 ,
78 Vasile Cîrtoaje
a12 + an−1
2
− 2t 2 = 2(t 2 − a1 an−1 ),
1
a14 + an−1
4
− 2t 4 = (t 2 − a1 an−1 )(7a12 + 7an−1 2
+ 10a1 an−1 ),
2
we may write the inequality F (a1 , a2 , . . . , an−2 , an−1 , an ) ≥ F (t, a2 , . . . , an−2 , t, an ) as
follows:
(a12 + an−1
2
− 2t 2 )(2t 2 + a12 + 2a22 + · · · + 2an−2
2
+ an−1
2
+ 2an2 ) ≥ k(a14 + an−1
4
− 2t 4 ),
Since
2an2 ≥ a12 + an−1
2
,
it suffices to show that
8 t 2 + a12 + an−1
2
≥ k(7a12 + 7an−1
2
+ 10a1 an−1 2).
Since 2d 2 ≥ a2 + c 2 , we have
and
3a4 + d 4 − 4 = 12(a − 1)2 (7a2 − 22a + 21),
the desired inequality can be written in the obvious form
which is equivalent to
Since
9(a2 + c 2 ) − 42ac ≥ −6(a + c)2 = −24t 2 ,
it suffices to show that
(d − t)(11d + 7t) ≥ 0.
Since
3a2 + d 2 − 4 = 12(a − 1)2 ,
11 72a2 − 144a + 107
3a2 + d 2 + =
6 6
and
3a4 + d 4 − 4 = 12(a − 1)2 (7a2 − 22a + 21),
the desired inequality can be written in the obvious form
5 7
The equality holds for a = b = c = d = 1, and also for a = b = c = and d =
9 3
(or any cyclic permutation).
82 Vasile Cîrtoaje
2
9 522
20(a2 + c 2 ) − 8ac + 64d 2 + 6b2 + 7b − (a + c + d) − (a + c + d)2 ≥ 0.
7 49
The inequality is true if
522
20(a2 + c 2 ) − 8ac + 64d 2 − (a + c + d)2 ≥ 0,
49
which is equivalent to
261
10(a2 + c 2 ) − 4ac + 32d 2 − (a + c + d)2 ≥ 0.
49
Arithmetic Mean Method and Arithmetic Compensation Method 83
Since
10(a2 + c 2 ) − 4ac ≥ 4(a + c)2 = 16t 2
and
261 16
< ,
49 3
it suffices to show that
16
16t 2 + 32d 2 − (2t + d)2 ≥ 0,
3
which is equivalent to the obvious inequality
(d − t)(5d + t) ≥ 0.
Since
3a2 + d 2 − 4 = 12(a − 1)2 ,
4
F (a, b, c, d) = (a2 + b2 + c 2 + d 2 + 4)2 − 64 − (a4 + b4 + c 4 + d 4 − 4).
3
Assume that a ≤ b ≤ c ≤ d and show that
a+c
F (a, b, c, d) ≥ F (t, b, t, d), t= .
2
Using the identities
1
a2 + c 2 − 2t 2 = 2(t 2 − ac), a4 + c 4 − 2t 4 = (t 2 − ac)(7a2 + 7c 2 + 10ac),
2
we may write the inequality F (a, b, c, d) ≥ F (t, b, t, d) as
4
(a2 + c 2 − 2t 2 )(2t 2 + a2 + 2b2 + c 2 + 2d 2 + 8) − (a4 + c 4 − 2t 4 ) ≥ 0,
3
2
2(t 2 − ac)(2t 2 + a2 + 2b2 + c 2 + 2d 2 + 8) − (t 2 − ac)(7a2 + 7c 2 + 10ac) ≥ 0,
3
(t 2 − ac)(−5a2 − 5c 2 − 14ac + 12b2 + 12d 2 + 48) ≥ 0.
This inequality is true if
which is equivalent to
which reduces to
4(a − c)2 ≥ 0.
By the AM-Corollary, it suffices to prove the original inequality for a = b = c. That
is, to show that 3a + d = 4 involves
Since
3a2 + d 2 − 4 = 12(a − 1)2 ,
Arithmetic Mean Method and Arithmetic Compensation Method 85
3a2 + d 2 + 12 = 4(3a2 − 6a + 7)
and
3a4 + d 4 − 4 = 12(a − 1)2 (7a2 − 22a + 21),
the desired inequality can be written in the obvious form
a(a − 1)2 (a + 2) ≥ 0.
1
a2 + c 2 − 2t 2 = 2(t 2 − ac), a4 + c 4 − 2t 4 = (t 2 − ac)(7a2 + 7c 2 + 10ac),
2
we may write the inequality F (a, b, c, d) ≥ F (t, b, t, d) as follows:
2 32
12 4
a + c 2 − 2t 2 2t 2 + a2 + 2b2 + c 2 + 2d 2 + − a + c 4 − 2t 4 ≥ 0,
11 11
32
22(t 2 − ac) 2t 2 + a2 + 2b2 + c 2 + 2d 2 + − 6(t 2 − ac)(7a2 + 7c 2 + 10ac) ≥ 0,
11
(t 2 − ac)(−9a2 − 9c 2 − 38ac + 44b2 + 44d 2 + 64) ≥ 0.
This inequality is true if
which reduces to
2(a2 + c 2 ) + 15(a − c)2 ≥ 0.
By the AM-Corollary, it suffices to prove the original inequality for a = b = c. That
is, to show that 3a + d = 4 involves
(3a2 + d 2 − 4) 33a2 + 11d 2 + 76 ≥ 12 3a4 + d 4 − 4 .
Since
3a2 + d 2 − 4 = 12(a − 1)2 ,
33a + 11d 2 + 76 = 12(11a2 − 22a + 21)
2
and
3a4 + d 4 − 4 = 12(a − 1)2 (7a2 − 22a + 21),
the desired inequality can be written in the obvious form
a2 (a − 1)2 ≥ 0.
a1 + a2 + · · · + an = m, m ∈ {1, 2, . . . , n},
then
1 1 1 m
+ + ··· + ≥ n− .
1 + a12 1 + a22 1 + an2 2
(Vasile C., 2005)
1 1 1 m
F (a1 , a2 , . . . , an ) = + + ··· + −n+ .
1 + a12 1 + a22 1 + an2 2
Assume that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
Arithmetic Mean Method and Arithmetic Compensation Method 87
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
Then, by AC-Corollary, we have
where
k m m(m − k)2
f (k) = n − k + −n+ =
1 + (m/k) 2 2 2(m2 + k2 )
m
is the value of F for a1 = · · · = an−k = 0 and an−k+1 = · · · = an = . Obviously, we
k
have f (k) ≥ 0 (with equality for k = m), therefore F (a1 , a2 , . . . , an ) ≥ 0.
To prove that (*) implies (**), we write (*) as
1 1 2
+ − < 0,
1 + a12 1 + a22 1 + t 2
2 − 2a1 a2 − a1 a2 (a1 + a2 )2 ≥ 0.
where
k 16 12(k − 3)(k − 4)
f (k) = 4 − k + − =
1 + 3(2/k)2 7 7(k2 + 12)
is the value of F when 4 − k of the numbers a, b, c, d are zero, and the other k
2
numbers are equal to . Obviously, we have f (k) ≥ 0 for k ∈ {1, 2, 3, 4} (with
k
equality for k = 3 and k = 4), therefore F (a1 , a2 , . . . , an ) ≥ 0.
To prove that (*) implies (**), we write (*) as
1 1 2
+ − < 0,
1 + 3a2 1 + 3b2 1 + 3t 2
(t 2 − a b)(6t 2 − 1 + 3a b)
< 0,
(1 + 3a2 )(1 + 3b2 )(1 + 3t 2 )
6t 2 < 1 − 3a b, (A)
and (**) as
1 1 1
+ −1− ≥ 0,
1 + 3a2 1 + 3b2 1 + 3(a + b)2
a b(1 − 3a b − 18a bt 2 )
≥ 0.
(1 + 3a2 )(1 + 3b2 )(1 + 12t 2 )
For the nontrivial case a b = 0, this is true if
1
6t 2 ≤ − 1.
3a b
Arithmetic Mean Method and Arithmetic Compensation Method 89
1
1 − 3a b ≤ − 1,
3a b
which is equivalent to
(3a b − 1)2 ≥ 0.
1 2
The equality holds for a = b = c = d = , and also for a = 0 and b = c = d =
2 3
(or any cyclic permutation).
Assume that
F (a, b, c, d) < F (t, t, c, d) (*)
involves
F (a, b, c, d) ≥ F (0, a + b, c, d), (**)
where
a+b
t= , a = b.
2
By the AC-Corollary, it suffices to prove that
4 4 4
F (1, 1, 1, 1) ≥ 0, F 0, , , ≥ 0,
3 3 3
F (0, 0, 2, 2) ≥ 0, F (0, 0, 0, 4) ≥ 0.
We have
4 4 4 2600
F (1, 1, 1, 1) = 9, F 0, , , = ,
3 3 3 729
F (0, 0, 2, 2 ) = 0, F (0, 0, 0, 4) = 8.
To prove that (*) implies (**), we write (*) as
(1 + t 2 )2 − (1 + a2 )(1 + b2 ) (1 + c 2 )(1 + d 2 ) < 0,
(1 + t 2 )2 − (1 + a2 )(1 + b2 ) < 0,
90 Vasile Cîrtoaje
(t 2 − a b)(t 2 + a b − 2) < 0,
2 − a b > t 2, (A)
and (**) as
1 + (a + b)2 − (1 + a2 )(1 + b2 ) (1 + c 2 )(1 + d 2 ) ≥ 0,
1 + (a + b)2 − (1 + a2 )(1 + b2 ) ≥ 0,
a b(2 − a b) ≥ 0.
This inequality is true if 2 − a b ≥ 0. Using (A), we have
2 − a b > t 2 ≥ 0.
2
P 1.48. If a1 , a2 , . . . , an are nonnegative real numbers such that a1 + a2 +· · ·+ an = ,
3
then ai a j 1
≤ .
1≤i≤ j≤n
(1 − a i )(1 − a j ) 4
Assume that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
92 Vasile Cîrtoaje
Then, by AC-Corollary, we only need to prove the original inequality for the case
when n−k of the variables a1 , a2 , . . . , an are zero and the other k variables are equal
2
to , where k ∈ {1, 2, · · · , n}; that is, to show that
3k
2 2
k 3k 1
≤ ,
2 1− 2 2 4
3k
which is equivalent to
2k(k − 1) 1
≤ ,
(3k − 2)2 4
(k − 2)2 ≥ 0.
To prove that (*) implies (**), we write the inequality (*) as
n
t2 a1 a2 2t a1 a2 ai
− + − − < 0,
(1 − t) 2 (1 − a1 )(1 − a2 ) 1 − t 1 − a1 1 − a2 i=3 1 − ai
t 2 − a1 a2 n
ai
1 − 2t − 2(1 − t) < 0,
(1 − a1 )(1 − a2 )(1 − t)2 i=3
1 − ai
n
ai
1 − 2t − 2(1 − t) < 0, (A)
i=3
1 − ai
and (**) as
n
−a1 a2 2t a1 a2 ai
+ − − ≥ 0,
(1 − a1 )(1 − a2 ) 1 − 2t 1 − a1 1 − a2 i=3 1 − ai
a1 a2 n
ai
1 − 2t − 2(1 − t) ≤ 0.
(1 − a1 )(1 − a2 )(1 − 2t) i=3
1 − ai
Clearly, this inequality follows immediately from (A), therefore (*) implies (**).
This completes the proof.
1
The equality holds when a1 = a2 = · · · = an−2 = 0 and an−1 = an = (or any
3
permutation).
Assume that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
Then, by AC-Corollary, we only need to prove the original inequality for the case
when n−k of the variables a1 , a2 , . . . , an are zero and the other k variables are equal
1
to , where k ∈ {1, 2, · · · , n}; that is, to show that
k
1 2
k k n
≥ ,
2 1− 1 2 2(n − 1)
k
n
ai
2t − 1 + 2(1 − t) <0
i=3
1 − ai
and
a1 a2 n
ai
2t − 1 + 2(1 − t) ≤ 0,
(1 − a1 )(1 − a2 )(1 − 2t) i=3
1 − ai
If
F (a, b, c, d) < F (t, t, c, d) (*)
involves
F (a, b, c, d) ≥ F (0, a + b, c, d), (**)
where
a+b
t= , a = b.
2
then, by the AC-Corollary, it suffices to prove that
F (0, 0, 0, 4) ≥ 0, F (0, 0, 2, 2) ≥ 0,
4 4 4
F 0, , , ≥ 0, F (1, 1, 1, 1) ≥ 0.
3 3 3
We have
F (0, 0, 0, 4) = 112, F (0, 0, 2, 2) = 76,
4 4 4
F 0, , , = 0, F (1, 1, 1, 1) = 0.
3 3 3
The inequality (*) is equivalent to
(1 + 3t)2 − (1 + 3a)(1 + 3b) (1 + 3c)(1 + 3d) − 131(t 2 − a b)cd < 0,
Assume that
F (a1 , a2 , a3 , . . . , an ) < F (t, t, a3 , . . . , an ) (*)
involves
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an ), (**)
where
a1 + a2
t= , a1 = a2 .
2
Then, by AC-Corollary, we only need to prove the original inequality for the case
when n−k of the variables a1 , a2 , . . . , an are zero and the other k variables are equal
s
to , where k ∈ {1, 2, · · · , n}; that is, to show that
k
s 2 3 3
k s 4n(n − 2) k s
k +2 + ≥ 2 ,
k 3 k 3(n − 1)2 2 k
which is equivalent to
(n − 1)2 − 1 (k − 1)2 − 1
≥ ,
(n − 1)2 (k − 1)2
1 1
≥ .
(k − 1)2 (n − 1)2
The inequality (*) is equivalent to
−2a1 a2 + a1 a2 (a3 + · · · + an ) ≥ 0,
a1 a2 (2 − a3 − · · · − an ) ≤ 0.
This inequality follows from (A), therefore (*) involves (**).
2
The equality holds when a1 = a2 = · · · = an = .
n−1
From
F (a, b, c, d) − f (a + c, b, 0, d) = c(b − d)(a + c − b − d)
and
F (a, b, c, d) − f (0, b, a + c, d) = a(d − b)(c + a − d − b),
we get
Similarly, we have
Therefore, f (a, b, c, d) is maximum when one of a and c is zero, and one of b and d
is zero. Thus, it suffices to consider
one of these cases. For instance, for c = d = 0,
we need to show that a + b = 3 involves
a b(a + 2b) ≤ 2.
We have
pqr Method
If P(a, b, c) is a symmetric polynomial having the degree not very large (practically,
less than or equal to eight), then the inequality can be written in the polynomial
form
P1 (p, q, r) ≥ 0,
where
p = a + b + c, q = a b + bc + ca, r = a bc.
In addition, the pqr method can be applied to prove inequalities where P1 (p, q, r)
is not a polynomial function.
The pqr method enables to prove an inequality P1 (p, q, r) ≥ 0 using the theorems
below.
Theorem 1 (see P 2.53 in Volume 1). If a ≥ b ≥ c are real numbers such that
a + b + c = p, a b + bc + ca = q,
where p and q are fixed real numbers satisfying p2 ≥ 3q, then the product r = a bc is
minimum for a = b, and maximum for b = c.
where p and q are fixed real numbers satisfying p2 ≥ 3q, then the product r = a bc is
minimum for a = b or c = 0, and maximum for b = c.
99
100 Vasile Cîrtoaje
2C a3 b + 2D a b3 = (C + D) a b(a2 + b2 ) + (C − D) a b(a2 − b2 )
= (C + D)(p q − 2q − pr) − (C − D)p(a − b)(b − c)(c − a),
2 2
|(a − b)(b − c)(c − a)| = (a − b)2 (b − c)2 (c − a)2
4(p2 − 3q)3 − (2p3 − 9pq + 27r)2
= , (*)
27
and
α ≥ 0, x ≥ 0, x 2 ≥ y 2,
then
βy +α x2 − y2 ≤ x α2 + β 2 ,
with equality for
βx = y α2 + β 2 .
Proof. Since
x α2 + β 2 − β y ≥ |β|x − β y ≥ |β|| y| − β y ≥ 0,
we can write the desired inequality as follows:
α x2 − y2 ≤ x α2 + β 2 − β y,
2
α2 (x 2 − y 2 ) ≤ x α2 + β 2 − β y ,
2
βx − y α2 + β 2 ≥ 0.
pqr Method 101
1
α=
, x = 2(p2 − 3q)3/2 , y = 2p3 − 9pq + 27r,
27
we get the following theorem:
Theorem 4. Let a, b, c be real numbers. For any real β, the following inequality holds
1 3/2
27β r + |(a − b)(b − c)(c − a)| ≤ 9β pq − 2β p3 + 2 + β 2 p2 − 3q ,
27
2.2 Applications
a + b + c > 0, a b + bc + ca ≥ 0,
then
(a) 3(a3 + b3 + c 3 ) ≥ (a + b + c)(a2 + b2 + c 2 );
15
(b) a3 + b3 + c 3 + a bc ≥ 0;
4
(c) 4(a3 + b3 + c 3 ) + 15a bc ≥ 3(a + b + c)(a b + bc + ca).
then
2(a3 + b3 + c 3 ) ≥ a b(a + b) + bc(b + c) + ca(c + a).
then
8(a3 + b3 + c 3 ) + 39a bc ≥ 7(a + b + c)(a b + bc + ca).
a + b + c > 0, a b + bc + ca ≥ 0,
then
11
(a) a4 + b4 + c 4 + a bc(a + b + c) ≥ 0;
4
(b) 4(a4 + b4 + c 4 ) + a bc(a + b + c) ≥ 5(a2 b2 + b2 c 2 + c 2 a2 ).
104 Vasile Cîrtoaje
a + b + c > 0, a b + bc + ca ≥ 0,
then
81(a3 + b3 + c 3 )2 ≥ 25(a2 + b2 + c 2 )3 .
a + b + c > 0, a b + bc + ca ≥ 0,
then
a2 (a − 2b + c) + b2 (b − 2c + a) + c 2 (c − 2a + b) ≥ 0.
a + b + c > 0, a b + bc + ca ≥ 0,
then
(a) 2(a3 + b3 + c 3 ) + 7a bc ≥ (a − b)(b − c)(c − a);
(b) a3 + b3 + c 3 + 2a bc ≥ a b2 + bc 2 + ca2 ;
! !
(c) 9(a3 + b3 + c 3 ) + 12a bc ≥ 2 a2 b + 11 a b2 .
a + b + c > 0, a b + bc + ca ≥ 0,
then
11
(a) a3 + b3 + c 3 + a bc ≥ (a − b)(b − c)(c − a);
4
! !
(b) 4(a3 + b3 + c 3 ) + 5a bc + 2 a2 b ≥ 6 a b2 ;
! !
(c) 36(a3 + b3 + c 3 ) + 30a bc + 13 a2 b ≥ 59 a b2 .
a + b + c > 0, a b + bc + ca ≥ 0,
then
1
a3 + b3 + c 3 − a bc ≥ 2(a − b)(b − c)(c − a).
4
pqr Method 105
a + b + c > 0, a b + bc + ca ≥ 0,
then
3 3
(b) a3 + b3 + c 3 − 3a bc ≥ (a − b)(b − c)(c − a).
2
then
a2 (a − b) + b2 (b − c) + c 2 (c − a) ≥ 0.
then
a2 (a + 2b − 3c) + b2 (b + 2c − 3a) + c 2 (c + 2a − 3b) ≥ 0.
a3 (a − 2b + c) + b3 (b − 2c + a) + c 3 (c − 2a + b) ≥ 0.
15
a4 + b4 + c 4 − a bc(a + b + c) ≥ (a + b + c)(a − b)(b − c)(c − a).
2
2(a b + bc + ca) ≥ a2 + b2 + c 2 ,
then
39
a4 + b4 + c 4 − a bc(a + b + c) ≥ (a + b + c)(a − b)(b − c)(c − a).
2
a4 + b4 + c 4 + 2a bc(a + b + c) ≥ a b3 + bc 3 + ca3 .
a4 + b4 + c 4 + 2 (a3 b + b3 c + c 3 a) ≥ 0.
(a2 + b2 + c 2 )2 + (1 + 2 7 )(a3 b + b3 c + c 3 a) ≥ 0.
a4 + b4 + c 4 ≥ (2 3 − 1)(a b3 + bc 3 + ca3 ).
a4 + b4 + c 4 + (1 + 2 3 )(a3 b + b3 c + c 3 a) ≥ 0.
a4 + b4 + c 4 + 2 2 (a3 b + b3 c + c 3 a) ≥ a b3 + bc 3 + ca3 .
(a + b + c)(a3 + b3 + c 3 ) + 5(a3 b + b3 c + c 3 a) ≥ 0.
1+ 21 + 8 7
k(a b + bc + ca) = a + b + c ,
2 2 2
k≥ ≈ 3.7468,
2
then
a3 b + b3 c + c 3 a ≥ 0.
a4 + b4 + c 4 + 5(a3 b + b3 c + c 3 a) ≥ 6(a2 b2 + b2 c 2 + c 2 a2 ).
a2 b2 c 2
+ + + 5(a b + bc + ca) ≥ 18.
b c a
a b c 5(a2 + b2 + c 2 − a b − bc − ca)
+ + −3≥ .
b c a a2 + b2 + c 2 + a b + bc + ca
a b c 16(a2 + b2 + c 2 − a b − bc − ca)
+ + −3≥ 2 .
b c a a + b2 + c 2 + 6(a b + bc + ca)
a + b + c = 3, a b + bc + ca ≥ 0,
then
a3 b + b3 c + c 3 a + 18 3 ≥ a b3 + bc 3 + ca3 .
81 2
a3 b + b3 c + c 3 a + ≥ a b3 + bc 3 + ca3 .
32
pqr Method 109
2.3 Solutions
Solution. Let
p = a + b + c, q = a b + bc + ca, r = a bc.
which is equivalent to
A4 − A2 q + 3kq2 ≥ 0,
where
A4 = (a2 + 2bc)[a2 + k(k + 1)bc], A2 = (k + 1) a2 + k(k + 3) bc.
We have
a4 + b4 + c 4 = (a2 + b2 + c 2 )2 − 2(a2 b2 + b2 c 2 + c 2 a2 )
= (p2 − 2q)2 − 2(q2 − 2pr)
= p4 − 4p2 q + 2q2 + 4pr,
therefore
A4 = a4 + (k2 + k + 2)a bc a + 2k(k + 1) b2 c 2
= p − 4p q + 2q + 4pr + (k + k + 2)pr + 2k(k + 1)(q2 − 2pr)
4 2 2 2
For given p and q, the left hand side of the desired inequality A4 − A2 q + 3kq2 ≥ 0
is a linear function of r. Therefore, this function is minimum when r is either
minimum or maximum. Thus, according to Theorem 1, it suffices to prove the
original inequality for b = c. In this case, the inequality becomes
(a − b)2 (a − k b)2 ≥ 0.
a
The equality holds for a = b = c, and also for = b = c (or any cyclic permuta-
k
tion). If k = 0, then the equality holds for a = b = c, and also for a = 0 and b = c
(or any cyclic permutation).
110 Vasile Cîrtoaje
a + b + c > 0, a b + bc + ca ≥ 0,
then
(a) 3(a3 + b3 + c 3 ) ≥ (a + b + c)(a2 + b2 + c 2 );
15
(b) a3 + b3 + c 3 + a bc ≥ 0;
4
(c) 4(a3 + b3 + c 3 ) + 15a bc ≥ 3(a + b + c)(a b + bc + ca).
(Vasile C., 2006)
2p3 − 7pq + 9r ≥ 0,
where
p = a + b + c, q = a b + bc + ca, r = a bc.
According to Theorem 1, for fixed p and q, the product r is minimum when
a = b ≥ c.
a + b + c = 2a + c, a b + bc + ca = a(a + 2c),
a ≥ c, 2a + c > 0, a + 2c ≥ 0
involve
3(2a3 + c 3 ) ≥ (2a + c)(2a2 + c 2 ),
which is equivalent to
(a − c)2 (a + c) ≥ 0.
This inequality is true because
15
3r + p3 − 3pq + r ≥ 0,
4
pqr Method 111
then
2(a3 + b3 + c 3 ) ≥ a b(a + b) + bc(b + c) + ca(c + a).
(Vasile C., 2006)
2p3 − 7pq + 9r ≥ 0.
According to Theorem 1, it suffices to prove the original inequality for a = b ≥ c.
Since
a ≥ c, 2a + c > 0, a+c ≥0
involve
2(2a3 + c 3 ) ≥ 2a(a2 + ac + c 2 ),
which is equivalent to
a3 − a2 c − ac 2 + c 3 ≥ 0,
(a + c)(a − c)2 ≥ 0.
The equality holds for a = b = c > 0, and also for a = b = −c > 0 (or any cyclic
permutation).
which is true if
(2a + b + c)(a + b + 2c) ≥ (a + c)2 .
This inequality is equivalent to the hypothesis a2 + b2 + c 2 + 3(a b + bc + ca) ≥ 0.
Case 2: a + b + 2c < 0. This case is not possible because, as shown at the preceding
case 1, the hypothesis a2 + b2 + c 2 + 3(a b + bc + ca) ≥ 0 involves
(a + c)2
a + b + 2c ≥ ≥ 0.
2a + b + c
then
8(a3 + b3 + c 3 ) + 39a bc ≥ 7(a + b + c)(a b + bc + ca).
(Vasile C., 2006)
First Solution. Write the inequality as
a ≥ c, a + 4c ≥ 0
involve
8(2a3 + c 3 ) + 39a2 c ≥ 7(2a + c)(a2 + 2ac),
which is equivalent to
a3 + 2a2 c − 7ac 2 + 4c 3 ≥ 0,
(a + 4c)(a − c)2 ≥ 0.
The equality holds for a = b = c > 0, and also for a = b = −4c > 0 (or any cyclic
permutation).
114 Vasile Cîrtoaje
a + b + c > 0, a b + bc + ca ≥ 0,
then
11
(a) a4 + b4 + c 4 + a bc(a + b + c) ≥ 0;
4
(b) 4(a4 + b4 + c 4 ) + a bc(a + b + c) ≥ 5(a2 b2 + b2 c 2 + c 2 a2 ).
(Vasile C., 2006)
Solution. We have
a4 + b4 + c 4 = (a2 + b2 + c 2 )2 − 2(a2 b2 + b2 c 2 + c 2 a2 )
= (p2 − 2q)2 − 2(q2 − 2pr)
= p4 − 4p2 q + 2q2 + 4pr.
27
p4 − 4p2 q + 2q2 + pr ≥ 0.
4
According to Theorem 1, it suffices to prove the original inequality for a = b ≥ c.
Clearly, the original inequality is true for a = b ≥ c ≥ 0. Consider further that
a = b > 0 > c. Since
a + b + c = 2a + c, a b + bc + ca = a(a + 2c),
involve
11 2
2a4 + c 4 + a c(2a + c) ≥ 0,
4
which is equivalent to
(a + 2c)(8a3 + 6a2 c − ac 2 + 2c 3 ) ≥ 0.
The last inequality is true since a + 2c ≥ 0 and
The equality holds for a = b = −2c > 0 (or any cyclic permutation).
pqr Method 115
a ≥ c, 2a + c > 0, a + 2c ≥ 0
involve
4(2a4 + c 4 ) + a2 c(2a + c) ≥ 5(a4 + 2a2 c 2 ),
which is equivalent to
Remark. The inequality in (b) is sharper than the inequalities in (a) because this
last inequality can be obtained by summing the inequality in (b) to the obvious
inequalities
(a b + bc + ca)2 ≥ 0.
a + b + c > 0, a b + bc + ca ≥ 0,
then
81(a3 + b3 + c 3 )2 ≥ 25(a2 + b2 + c 2 )3 .
(Vasile C., 2006)
27
p4 − 4p2 q + 2q2 + pr ≥ 0.
4
116 Vasile Cîrtoaje
Case 1: c < 0. Due to homogeneity, we may set c = −1. So, we need to show that
a−2≥0
involves
81(2a3 − 1)2 ≥ 25(2a2 + 1)3 ,
which is equivalent to
62a6 − 150a4 − 162a3 − 75a2 + 28 ≥ 0.
Since
62a6 − 150a4 − 162a3 ≥ 248a4 − 150a4 − 162a3
= 98a4 − 162a3 ≥ 196a3 − 162a3 = 34a3 ,
it suffices to show that
34a3 − 75a2 + 28 ≥ 0, .
which is equivalent to
(a − 2)(34a2 − 7a − 14) ≥ 0.
This is true because
34a2 − 7a − 14 > 7a2 − 7a − 14 = 7(a + 1)(a − 2) ≥ 0.
The equality holds for a = b = −2c > 0 (or any cyclic permutation).
pqr Method 117
a + b + c > 0, a b + bc + ca ≥ 0,
then
a2 (a − 2b + c) + b2 (b − 2c + a) + c 2 (c − 2a + b) ≥ 0.
(Vasile C., 2006)
Solution. Let
p = a + b + c, q = a b + bc + ca, r = a bc.
p2 1
p = 1, 0≤q≤ = .
3 3
2 − 7q
≥ −3r + |(a − b)(b − c)(c − a)|.
3
Applying Corollary 1 for
−1
β= ,
9
we have
2 4
−3r + |(a − b)(b − c)(c − a)| ≤ −q + + (1 − 3q)3/2 ,
9 9
with equality for
2 − 7q 2 4
≥ −q + + (1 − 3q)3/2 ,
3 9 9
118 Vasile Cîrtoaje
which is equivalent to
1 − 3q ≥ (1 − 3q)3/2 ,
(1 − 3q)(1 − 1 − 3q) ≥ 0,
3q(1 − 3q)
≥ 0.
1+ 1 − 3q
The last inequality is true, with equality for q = 1/3 and q = 0. According to (*),
q = 1/3 involves r = 1/27, and q = 0 involves r = −1/9. Thus, the original
inequality is an equality for a = b = c > 0, and also for
a b+ bc +ca = 0, (a+ b+c)3 +9a bc = 0, a+ b+c > 0, (a− b)(b−c)(c −a) < 0.
The last equality conditions are equivalent to the condition that a, b, c satisfy
9w3 − 9w2 + 1 = 0.
a + b + c > 0, a b + bc + ca ≥ 0,
then
(a) 2(a3 + b3 + c 3 ) + 7a bc ≥ (a − b)(b − c)(c − a);
(b) a3 + b3 + c 3 + 2a bc ≥ a b2 + bc 2 + ca2 ;
! !
(c) 9(a3 + b3 + c 3 ) + 12a bc ≥ 2 a2 b + 11 a b2 .
(Vasile C., 2006)
Solution. Denote
p = a + b + c, q = a b + bc + ca, r = a bc.
p2 1
p = 1, 0≤q≤ = .
3 3
−13
Applying Corollary 1 for β = , we have
27
−13q 26 28
−13r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
3 27 27
with equality for
−13q 26 28
2 − 6q ≥ + + (1 − 3q)3/2 ,
3 27 27
which is equivalent to
28 − 45q ≥ 28(1 − 3q)3/2 .
Using the substitution
t= 1 − 3q, 0 ≤ t ≤ 1,
the inequality becomes
28 − 15(1 − t 2 ) ≥ 28t 3 ,
13 + 15t 2 − 28t 3 ≥ 0.
We have
13 + 15t 2 − 28t 3 = (1 − t)(13 + 13t + 28t 2 ) ≥ 0,
with equality for t = 1. Notice that t = 1 involves q = 0, and (*) gives r = −1/7.
Thus, if p = 1, then the original inequality is an equality for q = 0, r = −1/7 and
(a − b)(b − c)(c − a) > 0. More general, the equality holds for
7w3 − 7w2 + 1 = 0.
120 Vasile Cîrtoaje
7w3 − 7w2 + 1 = 0.
pqr Method 121
−13q 26 28
−13r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 .
3 27 27
67q −13q 26 28
2− ≥ + + (1 − 3q)3/2 ,
9 3 27 27
which is equivalent to
1 − 3q ≥ (1 − 3q)3/2 ,
(1 − 3q)(1 − 1 − 3q ) ≥ 0,
3q(1 − 3q)
≥ 0.
1− 1 − 3q
The last inequality is true, with equality for q = 1/3 and q = 0. The original
inequality is an equality for a = b = c > 0, and also for
The last equality conditions are equivalent to the condition that a, b, c satisfy
7w3 − 71w2 + 1 = 0.
122 Vasile Cîrtoaje
Remark. The inequality (c) is sharper than the inequalities (b), and the inequality
(b) is sharper than the inequalities (a). Indeed, the inequalities (b) can be obtained
by summing the inequality in (c) to the obvious inequality
and the inequalities (a) can be obtained by summing the inequality in (b) to the
inequality
1
(a + b + c)(a b + bc + ca) ≥ 0.
2
a + b + c > 0, a b + bc + ca ≥ 0,
then
11
(a) a3 + b3 + c 3 + a bc ≥ (a − b)(b − c)(c − a);
4
! !
(b) 4(a3 + b3 + c 3 ) + 5a bc + 2 a2 b ≥ 6 a b2 ;
! !
(c) 36(a3 + b3 + c 3 ) + 30a bc + 13 a2 b ≥ 59 a b2 .
(Vasile C., 2006)
Solution. Consider
p2 1
p = 1, 0≤q≤ = .
3 3
(a) Write the inequality as
11
3r + p3 − 3pq + r ≥ (a − b)(b − c)(c − a),
4
23
p3 − 3pq + r ≥ (a − b)(b − c)(c − a),
4
−23
1 − 3q ≥ r + (a − b)(b − c)(c − a).
4
It suffices to show that
−23r
1 − 3q ≥ + |(a − b)(b − c)(c − a)|.
4
−23
Applying Corollary 1 for β = , we have
108
−23 −23q 23 31
r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
4 12 54 54
pqr Method 123
31w3 − 31w2 + 4 = 0.
31w3 − 31w2 + 4 = 0.
The last equality conditions are equivalent to the condition that a, b, c satisfy
31w3 − 31w2 + 4 = 0.
Remark. The inequality (c) is sharper than the inequalities (b), and the inequality
(b) is sharper than the inequalities (a). Indeed, the inequalities (b) can be obtained
by summing the inequality in (c) to the obvious inequality
and the inequalities (a) can be obtained by summing the inequality in (b) to the
inequality
2(a + b + c)(a b + bc + ca) ≥ 0.
126 Vasile Cîrtoaje
a + b + c > 0, a b + bc + ca ≥ 0,
then
1
a3 + b3 + c 3 − a bc ≥ 2(a − b)(b − c)(c − a).
4
(Vasile C., 2006)
Solution. Assume
p2 1
p = 1, 0≤q≤ = ,
3 3
and write the inequality as
1
3r + p3 − 3pq − r ≥ 2(a − b)(b − c)(c − a),
4
11
p3 − 3pq + r ≥ 2(a − b)(b − c)(c − a),
4
−11
1 − 3q ≥ r + 2(a − b)(b − c)(c − a).
4
It suffices to show that
1 3 −11r
− q≥ + |(a − b)(b − c)(c − a)|.
2 2 8
−11
Applying Corollary 1 for β = , we have
216
−11 −11q 11 43
r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
8 24 108 108
with equality for
11 + 75t 2 − 86t 3 ≥ 0.
We have
11 + 75t 2 − 86t 3 = (1 − t)(11 + 11t + 86t 2 ) ≥ 0,
with equality for t = 1. Notice that t = 1 involves q = 0, and (*) gives r = −4/43.
Thus, if p = 1, then the original inequality is an equality for q = 0, r = −4/43 and
(a − b)(b − c)(c − a) > 0. More general, the equality holds for
43w3 − 43w2 + 4 = 0.
a + b + c > 0, a b + bc + ca ≥ 0,
then
3 3
(b) a3 + b3 + c 3 − 3a bc ≥ (a − b)(b − c)(c − a).
2
(Vasile C., 2006)
p2 1
p = 1, 0≤q≤ = .
3 3
2
|(a − b)(b − c)(c − a)| ≤
(1 − 3q)3/2 ,
3 3
with equality for
27r = 9q − 2. (*)
2 − (2 − 3)(1 − t 2 ) ≥ 2t 3 ,
3 + (2 − 3)t 2 − 2t 3 ≥ 0,
(1 − t)( 3 + 3 t + 2t 2 ) ≥ 0,
with equality for t = 1. Notice that t = 1 involves q = 0, and (*) gives r = −2/27.
Thus, if p = 1, then the original inequality is an equality for q = 0, r = −2/27 and
(a − b)(b − c)(c − a) > 0. More general, the equality holds for
that is,
b c
a=
=
>0
1+ 3 1− 3
(or any cyclic permutation).
(b) Write the inequality as
3 3
p3 − 3pq ≥ (a − b)(b − c)(c − a),
2
3 3
1 − 3q ≥ (a − b)(b − c)(c − a).
2
It suffices to show that
3 3
1 − 3q ≥ |(a − b)(b − c)(c − a)|.
2
pqr Method 129
1 − 3q ≥ (1 − 3q)3/2 .
which is equivalent to
(1 − 3q) 1 − 1 − 3q ≥ 0,
3q(1 − 3q)
≥ 0.
1+ 1 − 3q
The equality holds for q = 1/3 and q = 0. The original inequality is an equality for
a = b = c > 0, and also for
3 3
(a + b + c)(a b + bc + ca) ≥ 0.
2
then
a2 (a − b) + b2 (b − c) + c 2 (c − a) ≥ 0.
(Vasile C., 2006)
we get
−2 1
≤q≤ .
3 3
130 Vasile Cîrtoaje
then
a2 (a + 2b − 3c) + b2 (b + 2c − 3a) + c 2 (c + 2a − 3b) ≥ 0.
(Vasile C., 2006)
we get
1 1
≤q≤ .
5 3
Write the inequality as follows:
2(a3 + b3 + c 3 ) ≥ a b(a + b) − 5 a b(a − b),
which is equivalent to
3 (1 − 3q) ≥ 7 (1 − 3q)3/2 ,
132 Vasile Cîrtoaje
"
#
(1 − 3q) 3− 7(1 − 3q) ≥ 0.
This is true if
3 ≥ 7(1 − 3q),
4
which is equivalent to q ≥ . Indeed, we have
21
1 4
q≥ > .
5 21
The equality holds for a = b = c > 0.
Since the inequality remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume that p ≥ 0. For the nontrivial case p = 0, we may set
p = 1 (due to homogeneity). Thus, we need to show that
2 − 11q + 14q2
≥ −r + |(a − b)(b − c)(c − a).
3
Applying Corollary 1 for β = −1/27, we have
−q 2 4 7
−r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
3 27 27
with equality for
2 − 11q + 14q2 −q 2 4 7
≥ + + (1 − 3q)3/2 ,
3 3 27 27
which is equivalent to
which is equivalent to
(7q − 2)2 ≥ 0.
According to (*), q = 1/3 involves r = 1/27, and q = 2/7 involves r = −1/49.
Thus, the original inequality is an equality for a = b = c, and also for
The last equality conditions are equivalent to the condition that a, b, c satisfy
(a − b)(b − c)(c − a) ≤ 0
a3 (a − 2b + c) + b3 (b − 2c + a) + c 3 (c − 2a + b) ≥ 0.
2(p4 − 4p2 q + 2q2 + 4pr) ≥ q(p2 − 2q) − pr − 3p(a − b)(b − c)(c − a),
2p4 − 9p2 q + 6q2 ≥ −9pr − 3p(a − b)(b − c)(c − a).
Since the inequality remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume that p ≥ 0. For the nontrivial case p > 0, it suffices to
show that
2 − 9q + 6q2
≥ −3r + |(a − b)(b − c)(c − a).
3
Applying Corollary 1 for β = −1/9, we have
2 4
−3r + |(a − b)(b − c)(c − a)| ≤ −q + + (1 − 3q)3/2 ,
9 9
with equality for
2 − 9q + 6q2 2 4
≥ −q + + (1 − 3q)3/2 ,
3 9 9
which is equivalent to
2 − 9q + 9q2 ≥ 2(1 − 3q)3/2 ,
(1 − 3q)(2 − 3q) ≥ 2(1 − 3q)3/2 ,
(1 − 3q) 2 − 3q − 2 1 − 3q ≥ 0.
Since 1 − 3q ≥ 0, this is true if
The last equality conditions are equivalent to the condition that a, b, c satisfy
(a − b)(b − c)(c − a) ≤ 0
9w3 − 9w2 + 1 = 0;
Solution. Since the inequality remains unchanged by replacing a, b, c with −a, −b, −c,
respectively, we may assume that p ≥ 0. For the nontrivial case p > 0, it suffices to
show that
which is equivalent to
− 3 q 2 3 2 10
− 3 r + |(a − b)(b − c)(c − a)| ≤ + +
(1 − 3q)3/2 ,
3 27 3 3
with equality for
2 2 10
1 − 4q + 2q2 ≥ −q + + (1 − 3q)3/2 ,
9 9
136 Vasile Cîrtoaje
which is equivalent to
which is equivalent to
(2q + 1)2 ≥ 0.
According to (*), q = 1/3 involves r = 1/27, and q = −1/2 involves r = −1/3.
Thus, the original inequality is an equality for a = b = c, and also for
The last equality conditions are equivalent to the condition that a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
6w3 − 6w2 − 3w + 2 = 0.
2
a3 (a + b) + b3 (b + c) + c 3 (c + a) ≥ (a b + bc + ca)2 .
3
(Vasile C., 2006)
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For the nontrivial case p > 0, it suffices to show
that
2
2p4 − 7p2 q + q2 + 7pr ≥ p|(a − b)(b − c)(c − a)|.
3
Due to homogeneity, we may set p = 1. From p2 ≥ 3q, we get
1
q≤ .
3
Thus, we need to show that
2
2 − 7q + q2 ≥ −7r + |(a − b)(b − c)(c − a)|.
3
−7
Applying Corollary 1 for β = , we have
27
−7q 14 4 19
−7r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
3 27 27
with equality for
7(1 − 3q)3/2 = 19(9q − 27r − 2). (*)
2 −7q 14 4 19
2 − 7q + q2 ≥ + + (1 − 3q)3/2 ,
3 3 27 27
which is equivalent to
20 − 63q + 9q2 ≥ 2 19(1 − 3q)3/2 ,
(1 − 3q)(20 − 3q) ≥ 2 19(1 − 3q)3/2 ,
"
#
(1 − 3q) 20 − 3q − 2 19(1 − 3q) ≥ 0.
Since 1 − 3q ≥ 0, this is true if
which is equivalent to
(q + 6)2 ≥ 0,
According to (*), q = 1/3 involves r = 1/27, and q = −6 involves r = −7. Thus,
the original inequality is an equality for a = b = c, and also for
The last equality conditions are equivalent to the condition that a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
w3 − w2 − 6w + 7 = 0.
−5q 10 4 13
−5r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
3 27 27
pqr Method 139
13 2 −5q 10 4 13
1 − 5q + q ≥ + + (1 − 3q)3/2 ,
3 3 27 27
which is equivalent to
which is equivalent to
(13q − 3)2 ≥ 0,
According to (*), q = 1/3 involves r = 1/27, and q = 3/13 involves r = −1/169.
Thus, the homogeneous inequality is an equality for a = b = c, and also for
Remark 1. The inequalities in P 2.14, ... , P2.18 are particular cases of the following
generalization:
• Let A, B, C, D be real numbers such that
1 + A + B + C + D = 0, 3(1 + A) ≥ C 2 + C D + D2 .
140 Vasile Cîrtoaje
3(1 + A) = C 2 + C D + D2 .
7 + 2A + C + D − 3(2 + A − C − D)q ≥
≥2 (1 − A)2 − (1 − A)(C + D) + C 2 − C D + D2 · 1 − 3q.
Since
C 2 + C D + D2 (C + D)2
1+A= ≥ ,
3 4
we have
(C + D)2
2 + A − C − D = (1 + A) + 1 − C − D ≥ +1−C −D
4
(C + D − 2)2
= ≥ 0,
4
hence
7 + 2A + C + D − 3(2 + A − C − D)q ≥ 7 + 2A + C + D − (2 + A − C − D)
= (1 + A) + 4 + 2(C + D)
(C + D)2
≥ + 4 + 2(C + D)
4
(C + D + 4)2
= ≥ 0.
4
Thus, we only need to show that
[7 + 2A + C + D − 3(2 + A − C − D)q]2 ≥
4 (1 − A)2 − (1 − A)(C + D) + C 2 − C D + D2 (1 − 3q),
which is equivalent to
2
(2 + A − C − D)2 q − 4 − 5A + (1 + A)(C + D) + C 2 + D2 ≥ 0.
pqr Method 141
Remark 2. Replacing A, B, C, D with A/A0 , B/A0 , C/A0 , D/A0 , respectively, the state-
ment from Remark 1 becomes as follows:
• Let A0 , A, B, C, D be real numbers such that
15
a4 + b4 + c 4 − a bc(a + b + c) ≥ (a + b + c)(a − b)(b − c)(c − a).
2
(Vasile C., 2006)
15
a4 + b4 + c 4 − a bc(a + b + c) ≥ (a + b + c)|(a − b)(b − c)(c − a)|,
2
which is equivalent to
15
p4 − 4p2 q + 2q2 + 3pr ≥ p|(a − b)(b − c)(c − a)|.
2
1
0≤q≤ .
3
We need to show that
15 −6r
1 − 4q + 2q2 ≥
+ |(a − b)(b − c)(c − a)| .
2 15
−2
Applying Corollary 1 for β =
, we have
9 15
−6r −2q 4 14
+ |(a − b)(b − c)(c − a)| ≤
+
+
(1 − 3q)3/2 ,
15 15 9 15 9 15
142 Vasile Cîrtoaje
which is equivalent to
q(4q + 7) ≥ 0,
According to (*), q = 1/3 involves r = 1/27, and q = 0 involves r = −2/21. Thus,
the original inequality is an equality for a = b = c, and also for
The last equality conditions are equivalent to the condition that a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
21w3 − 21w2 + 2 = 0.
f (a, b, c) ≥ 0,
where
15 3
f (a, b, c) =a4 + b4 + c 4 − a bc(a + b + c) + a b− a b3
2
− k(a b + bc + ca)(a2 + b2 + c 2 − a b − bc − ca), k > 0,
where
15 15
A = k, B = k − 1, C = −k + , D = −k − .
2 2
We see that 1 + A + B + C + D = 0. According to the statement in Remark 1 from P
2.18, if
3(1 + A) = C 2 + C D + D2 ,
then the inequality holds for all real a, b, c. It is easy to show that this condition
is satisfied for k = 1/2. Since the inequality f (a, b, c) ≥ 0 for k = 1/2 is stronger
than the original inequality, the proof is completed.
2(a b + bc + ca) ≥ a2 + b2 + c 2 ,
then
39
a4 + b4 + c 4 − a bc(a + b + c) ≥ (a + b + c)(a − b)(b − c)(c − a).
2
(Vasile C., 2006)
Solution. Since the statement remains unchanged by replacing a, b, c with −a, −b, −c,
respectively, we may assume that p ≥ 0. For the nontrivial case p > 0, it suffices to
show that
39
a4 + b4 + c 4 − a bc(a + b + c) ≥ (a + b + c)|(a − b)(b − c)(c − a)|,
2
which is equivalent to
39
p4 − 4p2 q + 2q2 + 3pr ≥ p|(a − b)(b − c)(c − a)|.
2
Due to homogeneity, we may set p = 1. From p2 ≥ 3q and 2(a b + bc + ca) ≥
a2 + b2 + c 2 , we get
1 1
≤q≤ .
4 3
We need to show that
39 −6r
1 − 4q + 2q2 ≥
+ |(a − b)(b − c)(c − a)| .
2 39
144 Vasile Cîrtoaje
−2
Applying Corollary 1 for β =
, we have
9 39
−6r −2q 4 22
+ |(a − b)(b − c)(c − a)| ≤
+
+
(1 − 3q)3/2 ,
39 39 9 39 9 39
with equality for
which is equivalent to
(4q − 1)(q + 8) ≥ 0,
According to (*), q = 1/3 involves r = 1/27, and q = 1/4 involves r = 1/132.
Thus, the original inequality is an equality for a = b = c, and also for
The last equality conditions are equivalent to the condition that a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
f (a, b, c) ≥ 0,
where
15
f (a, b, c) =a4 + b4 + c 4 − a bc(a + b + c) + a3 b − a b3
2
−k 2 ab − a2 a2 − ab , k > 0,
pqr Method 145
2 2
= a2 − 3 ab a2 + 2 ab
= a4 + 4 a2 b2 + a bc a−3 a3 b − 3 a b3 ,
where
39 39
A0 = k + 1, A = 4k, B = k − 1, C = −3k + , D = −3k − .
2 2
We see that A0 + A + B + C + D = 0. According to the statement in Remark 2 from
P 2.18, if A0 > 0 and
3A0 (A0 + A) = C 2 + C D + D2 ,
then the inequality holds for all real a, b, c. It is easy to show that this condition
is satisfied for k = 3/4. Since the inequality f (a, b, c) ≥ 0 for k = 3/4 is stronger
than the original inequality, the proof is completed.
a4 + b4 + c 4 + 2a bc(a + b + c) ≥ a b3 + bc 3 + ca3 .
2(p4 − 4p2 q + 2q2 + 4pr) + 4pr ≥ q(p2 − 2q) − pr + p(a − b)(b − c)(c − a) ≥ 0,
2p4 − 9p2 q + 6q2 ≥ −13pr + p(a − b)(b − c)(c − a) ≥ 0,
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is clearly true. For
p > 0, it suffices to show that
1
q≤ .
3
Thus, we need to show that
−13
Applying Corollary 1 for β = , we have
27
−13q 26 28
−13r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
3 27 27
with equality for
−13q 26 28
2 − 9q + 6q2 ≥ + + (1 − 3q)3/2 ,
3 27 27
which is equivalent to
(a − b)(b − c)(c − a) ≥ 0
7w3 − 7w2 + 1 = 0.
a b c
= =
sin 2π
7
sin 4π
7 sin 8π
7
pqr Method 147
f (a, b, c) ≥ 0,
where
2 2
f (a, b, c) = a4 +b4 +c 4 +2a bc(a+b+c)−(a b3 +bc 3 +ca3 )− [p +m(p2 −3q)]2 , m ∈ ,
27
satisfies f (1, 1, 1) = 0. The inequality can be written as
a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ≥ 0,
a4 + b4 + c 4 + 2 (a3 b + b3 c + c 3 a) ≥ 0.
f (a, b, c) ≥ 0,
where
1+k
where
a4 + b4 + c 4 + k(a3 b + b3 c + c 3 a) ≥ 0
3 105 − 7
−1 ≤ k ≤ ≈ 1.4838,
16
3 105 − 7
where k = is a root of the equation 28 − 7k − 8k2 = 0. Actually, the
16
inequality holds for
−1 ≤ k ≤ k0 ,
where k0 ≈ 1.4894 is a root of the equation
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is clearly true. For
p > 0, it suffices to show that
1
q≤ .
3
Thus, we need to show that
2 − 9q + 10q2 1
≥ − r + |(a − b)(b − c)(c − a)| ≥ 0.
5 5
−1
Applying Corollary 1 for β = , we have
135
1 −5q 10 52
− pr + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
5 3 27 135
with equality for
2 − 9q + 10q2 −q 2 52
≥ + + (1 − 3q)3/2 ,
5 15 135 135
which is equivalent to
which is equivalent to
(t − 1)2 (15t 2 + 4t + 2) ≥ 0.
The last inequality is true, with equality for t = 1, that is q = 0. From (*), we get
r = −1/13. Thus, the original inequality is an equality when a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
150 Vasile Cîrtoaje
13w3 − 13w2 + 1 = 0.
f (a, b, c) ≥ 0,
where
2 2
f (a, b, c) = (a2 +b2 +c 2 )2 +2(a3 b+b3 c+c 3 a)−3(a b3 +bc 3 +ca3 )− [p +m(p2 −3q)]2
27
satisfies f (1, 1, 1) = 0. The inequality can be written as
a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ≥ 0,
3(1 + A) = C 2 + C D + D2 .
7 2
(p − 2q)2 + q(p2 − 2q) − pr − p(a − b)(b − c)(c − a) ≥ 0.
4
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is clearly true. For
p > 0, it suffices to show that
7 2
(p − 2q)2 + q(p2 − 2q) ≥ pr + p|(a − b)(b − c)(c − a)| ≥ 0.
4
pqr Method 151
7
(1 − 2q)2 + q(1 − 2q) ≥ r + |(a − b)(b − c)(c − a)|.
4
1
Applying Corollary 1 for β = , we have
27
q 2 4 7
r + |(a − b)(b − c)(c − a)| ≤ − + (1 − 3q)3/2 ,
3 27 27
with equality for
7 q 2 4 7
(1 − 2q)2 + q(1 − 2q) ≥ − + (1 − 3q)3/2 .
4 3 27 27
Using the substitution
t= 1 − 3q, t ≥ 0,
(*) turns into
which is equivalent to
2
3+ 7
t− 6( 7 − 2)t 2 + 2(3 − 7 )t + 1 ≥ 0.
2
3+ 7
The last inequality is true, with equality for t = , hence for
2
1 − t2 2+ 7
q= =− .
3 2
From (**), we get
3+ 7
r =−
.
4 7
152 Vasile Cîrtoaje
(a − b)(b − c)(c − a) ≥ 0
4 7+2 2
(p − 2q)2 ≥ q(p2 − 2q) − pr + p(a − b)(b − c)(c − a).
27
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is trivial. For p > 0, it
suffices to show that
4 7+2 2
(p − 2q)2 − q(p2 − 2q) ≥ −pr + p|(a − b)(b − c)(c − a)|.
27
Due to homogeneity, we may set p = 1. Thus, we need to show that
4 7+2
(1 − 2q)2 − q(1 − 2q) ≥ −r + |(a − b)(b − c)(c − a)|.
27
−1
Applying Corollary 1 for β = , we have
27
−q 2 4 7
−r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
3 27 27
with equality for
4 7+2 −q 2 4 7
(1 − 2q)2 − q(1 − 2q) ≥ + + (1 − 3q)3/2 ,
27 3 27 27
pqr Method 153
4 7 2
(1 − 2q)2 − (1 − 3q)3/2 − q(13 − 31q) ≥ 0.
27 27
Since q ≤ 0, this inequality is true if
4 7 2
(1 − 2q)2 − (1 − 3q)2 − q(13 − 31q) ≥ 0,
27 27
which is equivalent to
(a − b)(b − c)(c − a) ≥ 0
where
A = 2k − m + 2, B = 5k − m, C = k + m, D = k + m + 1 − 2 7.
According to the statement in Remark 2 from P 2.18, the inequality holds for all
real a, b, c if
1 + A+ B + C + D = 0
154 Vasile Cîrtoaje
and
3(1 + A) = C 2 + C D + D2 .
2( 7 − 2) 7 7−5
These condition are satisfied for k = , and m = . Thus, the proof
9 9
is completed.
(a2 + b2 + c 2 )2 + (1 + 2 7 )(a3 b + b3 c + c 3 a) ≥ 0.
4 7−2 2
(p − 2q)2 + q(p2 − 2q) − pr − p(a − b)(b − c)(c − a) ≥ 0.
27
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is clearly true. For
p > 0, it suffices to show that
4 7−2 2
(p − 2q)2 + q(p2 − 2q) ≥ pr + p|(a − b)(b − c)(c − a)| ≥ 0.
27
Due to homogeneity, we may set p = 1. From p2 ≥ 3q, we get
1
0≤q≤ .
3
Thus, we need to show that
4 7−2
(1 − 2q)2 + q(1 − 2q) ≥ r + |(a − b)(b − c)(c − a)|.
27
1
Applying Corollary 1 for β = , we have
27
q 2 4 7
r + |(a − b)(b − c)(c − a)| ≤ − + (1 − 3q)3/2 ,
3 27 27
with equality for
4 7−2 q 2 4 7
(1 − 2q)2 + q(1 − 2q) ≥ − + (1 − 3q)3/2 ,
27 3 27 27
which can be written as
4 7 2 2
(1 − 2q)2 − (1 − 3q)3/2 + 1 − (1 − 2q)2 + q(1 − 3q) ≥ 0.
27 27 3
This is true if
(1 − 2q)2 − (1 − 3q)3/2 ≥ 0.
Using the substitution
t= 1 − 3q, 0 ≤ t ≤ 1,
the inequality becomes
(1 + 2t 2 )2 ≥ 9t 3 ,
which is equivalent to
(1 − t)(1 + t + 5t 2 − 4t 3 ) ≥ 0.
Since
1 + t + 5t 2 − 4t 3 ≥ 1 + t + 2t 2 − 4t 3 = (1 − t)(1 + 2t + 4t 2 ) ≥ 0,
the inequality is true, with equality for t = 1, hence for q = 0. From (*), we get
1 1
r=
−2 .
27 7
Thus, the original inequality is an equality when a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
where
A = −2k + m + 2, B = −5k + m, C = −k − m, D = −k − m + 1 + 2 7.
According to the statement in Remark 1 from P 2.18, the inequality holds for all
real a, b, c if
1 + A+ B + C + D = 0
and
3(1 + A) = C 2 + C D + D2 .
2(2 + 7) 5+7 7
These conditions are satisfied for k = and m = . Thus, the
9 9
proof is completed.
(a + b + c)2 ≥ a2 + b2 + c 2 ,
where
E = a2 + b2 + c 2 − 7 (a b + bc + ca).
We can prove the inequalities (B) and (C) in a similar way. The original inequality
and the inequalities (A), (B) and (C) are equalities in the same conditions.
a4 + b4 + c 4 ≥ (2 3 − 1)(a b3 + bc 3 + ca3 ).
1 1+2 3
k=
= .
2 3−1 11
Write the inequality as follows:
2k(a4 + b4 + c 4 ) ≥ a b(a2 + b2 ) − a b(a2 − b2 ),
2k(p4 − 4p2 q + 2q2 + 4pr) ≥ q(p2 − 2q) − pr + p(a − b)(b − c)(c − a).
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is true. For p > 0, it
suffices to show that
2k(p4 − 4p2 q + 2q2 + 4pr) ≥ q(p2 − 2q) − pr + p|(a − b)(b − c)(c − a)|.
which is equivalent to
2k(1 − 4q + 2q2 ) − q(1 − 2q) ≥ −(1 + 8k)r + |(a − b)(b − c)(c − a)|.
−(1 + 8k)
Applying Corollary 1 for β = , we have
27
−(1 + 8k)r + |(a − b)(b − c)(c − a)| ≤
97w3 − 97w2 + 10 + 3 = 0.
Second Solution. We will find a stronger inequality of the form
A = 2k − m, B = 5k − m, C = k + m, D = k + m + 1 − 2 3.
According to the statement in Remark 1 from P 2.18, the inequality holds for all
real a, b, c if
1 + A+ B + C + D = 0
and
3(1 + A) = C 2 + C D + D2 .
2( 3 − 1) 7( 3 − 1)
These conditions are satisfied for k = and m = . Thus, the
9 9
proof is completed.
pqr Method 159
a4 + b4 + c 4 + (1 + 2 3 )(a3 b + b3 c + c 3 a) ≥ 0.
Solution. Denoting
2 3−1
k= ,
11
we may write the inequality as follows:
2k(a4 + b4 + c 4 ) + a b(a2 + b2 ) + a b(a2 − b2 ) ≥ 0,
2k(p4 − 4p2 q + 2q2 + 4pr) + q(p2 − 2q) − pr − p|(a − b)(b − c)(c − a)| ≥ 0.
1
p = 1, 0≤q≤ .
3
Thus, we need to show that
1 − 8k
Applying Corollary 1 for β = , we have
27
(1 − 8k)r + |(a − b)(b − c)(c − a)| ≤
which is equivalent to
3 1 − (1 − 3q)3/2 − 3q(4 − 3q) ≥ 0.
3(1 − t 3 ) − (1 − t 2 )(3 + t 2 ) ≥ 0,
which is equivalent to
t 2 (1 − t)(2 − t) ≥ 0.
This inequality is true, with equality for t = 1, hence for q = 0. From (*), we get
−(10 − 3)
r= .
97
Thus, the original inequality is an equality when a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
97w3 − 97w2 + 10 − 3 = 0.
where
A = −2k + m, B = −5k + m, C = −k − m, D = −k − m + 1 + 2 3.
According to the statement in Remark 1 from P 2.18, the inequality holds for all
real a, b, c if
1 + A+ B + C + D = 0
and
3(1 + A) = C 2 + C D + D2 .
2(1 + 3) 7(1 + 3)
These conditions are satisfied for k = and m = . Thus, the
9 9
proof is completed.
a4 + b4 + c 4 + 2 2 (a3 b + b3 c + c 3 a) ≥ a b3 + bc 3 + ca3 .
2(p4 − 4p2 q + 2q2 + 4pr) + (m − 1)[q(p2 − 2q) − pr] ≥ (m + 1)p(a − b)(b − c)(c − a).
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. For p = 0, the inequality is true. For p > 0, it
suffices to show that
2(p4 − 4p2 q + 2q2 ) + (m − 1)q(p2 − 2q) ≥ (m − 9)r + (m + 1)p|(a − b)(b − c)(c − a)|.
1
p = 1, 0≤q≤ .
3
Thus, we need to show that
which is equivalent to
we have
−(1 − t 2 )(2 + t 2 ) 2 t 2 (1 − t)2
B= + (1 − t 3 ) = ≥ 0.
3 3 3
The inequality mA + 2B ≥ 0 is an equality for q = 0. From (*), we get
2−9
r= .
7
Thus, the original inequality is an equality when a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
7w3 − 7w2 + 9 − 2 = 0.
pqr Method 163
where
A = −2k + m, B = −5k + m, C = −k − m + 2 2, D = −k − m − 1.
According to the statement in Remark 1 from P 2.18, the inequality holds for all
real a, b, c if
1 + A+ B + C + D = 0
and
3(1 + A) = C 2 + C D + D2 .
2 2 7 2
These conditions are satisfied for k = and m = . Thus, the proof is
9 9
completed.
(a + b + c)(a3 + b3 + c 3 ) + 5(a3 b + b3 c + c 3 a) ≥ 0.
p2 1
p = 1, 0≤q≤ = .
3 3
The desired inequality becomes as follows:
2(a + b + c)(a3 + b3 + c 3 ) + 5 a b(a2 + b2 ) + 5 a b(a2 − b2 ) ≥ 0,
2 − q − 10q2 −1
≥ r + |(a − b)(b − c)(c − a)|.
5 5
−1
Applying Corollary 1 for β = , we have
135
−1 −9q 2 52
r + |(a − b)(b − c)(c − a)| ≤ + + (1 − 3q)3/2 ,
5 135 135 135
with equality for
(1 − 3q)3/2 = 13(9q − 2 − 27r). (*)
Therefore, it suffices to show that
2 − q − 10q2 −9q 2 52
≥ + + (1 − 3q)3/2 ,
5 135 135 135
which can be rewritten as
13w3 − 13w2 + 1 = 0.
where
A = −2k + m, B = −5k + m, C = −k − m + 6, D = −k − m + 1.
According to the statement in Remark 1 from P 2.18, the inequality holds for all
real a, b, c if
1 + A+ B + C + D = 0
and
3(1 + A) = C 2 + C D + D2 .
8 28
These conditions are satisfied for k = and m = . Thus, the proof is completed.
9 9
1+ 21 + 8 7
k(a b + bc + ca) = a + b + c ,
2 2 2
k≥ ≈ 3.7468,
2
then
a3 b + b3 c + c 3 a ≥ 0.
(Vasile C., 2012)
1
q= .
k+2
Thus, we need to show that
1
Applying Corollary 1 for β = , we have
27
q 2 4 7
r + |(a − b)(b − c)(c − a)| ≤ − + (1 − 3q)3/2 ,
3 27 27
with equality for
q 2 4 7
q(1 − 2q) ≥ − + (1 − 3q)3/2 ,
3 27 27
which can be rewritten as
1 + 3(1 − t 2 ) − 3(1 − t 2 )2 ≥ 2 7 t 3 ,
1 + 3t 2 − 2 7 t 3 − 3t 4 ≥ 0,
1 + t + (2 + 7)t 2 1 − t + (2 − 7)t 2 ≥ 0.
We only need to show that
1 − t + (2 − 7)t 2 ≥ 0,
which is equivalent to
k − 1
k − 1
1 − ( 7 − 2) ≥ ,
k+2 k+2
(3 − 7)k + 7 ≥ (k − 1)(k + 2).
pqr Method 167
(2 7 − 5)k2 − (2 7 − 5)k − 3 ≤ 0,
k2 − k − 2 7 − 5 ≤ 0.
1 + 21 + 8 7
The last inequality is true for k ≥ , with equality for
2
1 + 21 + 8 7
k= .
2
From (*) and (**), we get the equality condition
which is equivalent to
5 1 5 + k − k2
14r = −1 + 5q − q2 = −1 + − = .
k + 2 (k + 2)2 (k + 2)2
1 + 21 + 8 7
k=
2
5 + k − k2 −1
r= =
.
14(k + 2)2 7(k + 2)2
1 1
w3 − w2 + w+
= 0.
k+2 7(k + 2)2
Solution. It suffices to show that there exist α, β, γ ∈ such that the sharper in-
equality f (a, b, c) ≥ 0 holds for any real a, b, c and f (1, 1, 1) = 0, where
f (a, b, c) = 3 a4 + 4 a3 b − 7 a b3 − bc(αa + β b + γc)2 ,
Solution. It suffices to show that there exist α, β, γ ∈ such that the sharper in-
equality f (a, b, c) ≥ 0 holds for any real a, b, c and f (1, 1, 1) = 0, where
f (a, b, c) = 16 a4 + 52 a3 b − 47 a b3 − bc(αa + β b + γc)2 ,
we have
f (a, b, c) = A0 a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ,
where
we choose
β = −2γ.
We have
we choose
γ=3
170 Vasile Cîrtoaje
and
β = −2γ = −6.
In addition, for
α= 21 − β − γ = 21 + 3
we have 3(A0 + A + B + C + D) = f (1, 1, 1) = 0. As a consequence, the inequality
2
16 a4 + 52 a3 b ≥ 47 a b3 + bc ( 21 + 3)a − 6b + 3c
holds for any real a, b, c. The original inequality is an equality for a = 0 and b = c/2
(or any cyclic permutation).
a4 + b4 + c 4 + 5(a3 b + b3 c + c 3 a) ≥ 6(a2 b2 + b2 c 2 + c 2 a2 ).
Solution. It suffices to show that there exist α, β, γ ∈ such that the sharper in-
equality f (a, b, c) ≥ 0 holds for any real a, b, c and f (1, 1, 1) = 0, where
f (a, b, c) = a4 + 5 a3 b − 6 a2 b2 − bc(αa + β b + γc)2 ,
we have
f (a, b, c) = a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ,
where
Choosing
3
γ=− ,
2
we have
9
A = −6 + 3β, B = −α(α + 2β − 3), C = 5 − β 2, D=− .
4
pqr Method 171
3(1 + A) ≥ C 2 + C D + D2 ,
It suffices to show that there exist α, β, γ ∈ such that the sharper inequality
f (a, b, c) ≥ 0 holds for any real a, b, c, where
f (a, b, c) = g(a, b, c) − bc(αa + β b + γc)2 ,
and
bc(αa + β b + γc)2 =2βγ b2 c 2 + α(α + 2β + 2γ)a bc a
+ β2 a 3 b + γ2 a b3 ,
we have
f (a, b, c) = A0 a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ,
where
A0 = 27, A = 162 − 2βγ, B = −149 − α(α + 2β + 2γ),
C = −148 − β , 2
D = 108 − γ2 .
Since
g(1, 1, 1) = g(0, 3, 1) = 0,
we need to have also
f (1, 1, 1) = f (0, 3, 1) = 0.
From
f (1, 1, 1) = −3(α + β + γ)2
and
f (0, 3, 1) = −3(3β + γ)2 ,
we get
γ = −3β, α = −β − γ = 2β,
therefore
we choose
β= 5, α = 2β = 2 5, γ = −α − β = −3 5.
As a consequence, the inequality
27(a + b + c)4 − 256(a3 b + b3 c + c 3 a) − 473a bc(a + b + c) ≥ 5 bc(2a + b − 3c)2
holds for any real a, b, c. The original inequality is an equality for a = b = c = 4/3,
and also for a = 0, b = 3 and c = 1 (or any cyclic permutation).
pqr Method 173
a2 b2 c 2
+ + + 5(a b + bc + ca) ≥ 18.
b c a
(Michael Rozenberg and Vasile C., 2009)
a2 b2 c 2 15(a b + bc + ca)
+ + + ≥ 6(a + b + c),
b c a a+b+c
a2 b2 a2 b2 30(a b + bc + ca)
+ + − + ≥ 12(a + b + c),
b a b a a+b+c
(p2 − 2q)q − pr p(a − b)(b − c)(c − a) 30q
+ + ≥ 12p,
r r p
30q
(p2 − 2q)q ≥ 13p − r − p(a − b)(b − c)(c − a).
p
It suffices to show that
30q
(p2 − 2q)q ≥ 13p − r + p|(a − b)(b − c)(c − a)|.
p
1 1 13 − 30q
Case 1: ≤ q ≤ . Applying Corollary 1 for β = , we have
9 3 27
(13 − 30q)q 2(13 − 30q) 2A
(13 − 30q)r + |(a − b)(b − c)(c − a)| ≤ − + (1 − 3q)3/2 ,
3 27 27
where
A= 27 + (13 − 30q)2 .
Therefore, it suffices to show that
(13 − 30q)q 2(13 − 30q) 2A
(1 − 2q)q ≥ − + (1 − 3q)3/2 ,
3 27 27
which is equivalent to
(1 − 3q) 13 − 36q − A 1 − 3q ≥ 0.
174 Vasile Cîrtoaje
This is true if
(13 − 36q)2 ≥ A2 (1 − 3q),
which can be written as
which is equivalent to
and
1 > 10q(1 − 2q).
The last inequality is true because
then
a2 b2 c 2
+ + + (k − 1)(a b + bc + ca) ≥ 3k.
b c a
a b c 5(a2 + b2 + c 2 − a b − bc − ca)
+ + −3≥ .
b c a a2 + b2 + c 2 + a b + bc + ca
(Vasile C., 2009)
19p2 − 39q
pq ≥ r + |(a − b)(b − c)(c − a)|.
p2 − q
1
0≤q≤ .
3
Thus, we need to show that
19 − 39q
q≥ r + |(a − b)(b − c)(c − a)|. (*)
1−q
176 Vasile Cîrtoaje
1 1 19 − 39q
Case 1: ≤ q ≤ . Applying Corollary 1 for β = , we have
15 3 27(1 − q)
This is true if
(19 − 54q)2 ≥ A2 (1 − 3q),
which can be written as
1
Case 2: 0 < q ≤ . Since
15
|(a − b)(b − c)(c − a)| = −27r 2 − 2(2p3 − 9pq)r + p2 q2 − 4q3
= −27r 2 − 2(2 − 9q)r + q2 − 4q3 ,
Ar 2 + 4Br + 4q3 ≥ 0,
where 2
19 − 39q
A = 27 + > 0,
1−q
(1 − 15q) + 24q2
B= > 0.
1−q
The original inequality is an equality for a = b = c.
a b c 16(a2 + b2 + c 2 − a b − bc − ca)
+ + −3≥ 2 .
b c a a + b2 + c 2 + 6(a b + bc + ca)
(Vasile C., 2012)
Solution. Write the inequality as follows:
a b a b 32(a2 + b2 + c 2 − a b − bc − ca)
+ + − −6≥ 2 .
b a b a a + b2 + c 2 + 6(a b + bc + ca)
1 1 41 − 60q
Case 1: ≤ q ≤ . Applying Corollary 1 for β = , we have
28 3 27(1 + 4q)
This is true if
(41 − 108q)2 ≥ A2 (1 − 3q),
which can be written as
Ar 2 + 4Br + 4q3 ≥ 0,
where 2
41 − 60q
A = 27 + > 0,
1+q
(1 − 21q) + 12q2
B= > 0.
1 + 4q
The original inequality is an equality for a = b = c, and also when a, b, c satisfy
a b c
π =
2π
=
4π
7 − tan 7 − tan 7 − tan
7 7 7
(or any cyclic permutation).
Remark. This inequality is stronger than the inequality
a b c 14(a2 + b2 + c 2 )
+ + +2≥ , (A)
b c a (a + b + c)2
a b c 7(a b + bc + ca) 17
+ + + ≥ , (B)
b c a a2 + b2 + c 2 2
because
16(a2 + b2 + c 2 − a b − bc − ca) 14(a2 + b2 + c 2 )
≥ −2
a2 + b2 + c 2 + 6(a b + bc + ca) (a + b + c)2
and
14(a2 + b2 + c 2 ) 17 7(a b + bc + ca)
−2≥ − ,
(a + b + c)2 2 a2 + b2 + c 2
180 Vasile Cîrtoaje
4π
7 − tan 7 − tan 7 − tan
7 7 7
(or any cyclic permutation).
p2 1
p = 1, 0≤q≤ = .
3 3
The desired inequality becomes as follows:
2 a2 a3 + 5 a b(a3 + b3 ) + 5 a b(a3 − b3 ) ≥ 0,
2 a2 a3 + 5 ab a3 − 5a bc a+5 a b(a3 − b3 ) ≥ 0,
2 a2 + 5 ab a3 − 5a bc a2 + 5 a b(a3 − b3 ) ≥ 0,
Since
A= 262 − 4q(331 − 211q) ≤ 26
and
1 − 3q ≤ 1,
it suffices to show that
which is equivalent to
3q(1 + 6q) ≥ 0.
For q = 0, (*) gives r = −1/13. Thus, the original inequality is an equality when
a, b, c satisfy
(a − b)(b − c)(c − a) ≥ 0
and are proportional to the roots of the equation
13w3 − 13w2 + 1 = 0.
a + b + c = 3, a b + bc + ca ≥ 0,
then
a3 b + b3 c + c 3 a + 18 3 ≥ a b3 + bc 3 + ca3 .
(Vasile C., 2008)
182 Vasile Cîrtoaje
0 ≤ q ≤ 3.
Since
4(p2 − 3q)3 − (2p3 − 9pq + 27r)2
|(a − b)(b − c)(c − a)| =
27
(p2 − 3q)3
≤2 = 2(3 − q) 3 − q,
27
r = q − 2,
a = −1, b = 2 − 2, c =2+ 2
81 2
a3 b + b3 c + c 3 a + ≥ a b3 + bc 3 + ca3 .
32
(Vasile C., 2008)
pqr Method 183
9 2 2
(a + b2 + c 2 )2 ≥ (a + b + c)(a − b)(b − c)(c − a). (*)
32
Since the statement remains unchanged by replacing a, b, c with −a, −b, −c, re-
spectively, we may assume p ≥ 0. Thus, it suffices to show that
9 2 2
(a + b2 + c 2 )2 ≥ (a + b + c)|(a − b)(b − c)(c − a)|,
32
which can be written as
9 2 2
(p − 2q)2 ≥ p|(a − b)(b − c)(c − a)|,
32
Since
4(p2 − 3q)3 − (2p3 − 9pq + 27r)2
|(a − b)(b − c)(c − a)| =
27
(p2 − 3q)3
≤2 ,
27
with equality for 27r = 9pq − 2p3 , we only need to show that
9 2 2
(p2 − 3q)3
(p − 2q) ≥ 2p
2
.
32 27
This inequality is equivalent to
3 −3 −7
p= , q= , r= ,
2 8 16
and
−3 −3 7
p= , q= , r= .
2 8 16
In the first case, a, b, c are the roots of the equation
16w3 − 24w2 − 6w + 7 = 0,
which is equivalent to
(2w − 1)(8w2 − 8w − 7) = 0.
In the second case, a, b, c are the roots of the equation
16w3 + 24w2 − 6w − 7 = 0,
which is equivalent to
(2w + 1)(8w2 + 8w − 7) = 0.
Thus,the original inequality is an equality for
1 2+3 2 2−3 2
a= , b= , c=
2 4 4
(or any cyclic permutation), and for
−1 −2 − 3 2 −2 + 3 2
a= , b= , c=
2 4 4
(or any cyclic permutation)
Chapter 3
x + y + z = p, x y + yz + z x = q,
where p and q are given real numbers satisfying p2 ≥ 3q, then the product r = x yz
is minimum and maximum when two of x, y, z are equal.
p = x + y + z, q = x y + yz + z x, r = x yz,
it can be rewritten as
185
186 Vasile Cîrtoaje
where A is the highest coefficient of f6 (x, y, z), and g1 (p, q) and g2 (p, q) are polyno-
mial functions of the form
are, respectively,
3, −3, −3, 3, 3, −3.
Therefore, the highest coefficient of f6 (x, y, z) is
A = 3(A1 − A2 − A3 + A4 + A5 − A6 + A7 ).
The polynomial
P1 (x, y, z) = (A1 x 2 + A2 yz)(B1 x 2 + B2 yz)(C1 x 2 + C2 yz)
Indeed, since
P1 (x, y, z) =A1 B1 C1 x 6 + A2 B2 C2 x3 y3
+ (B1 C1 A2 + C1 A1 B2 + A1 B1 C2 )x yz x3
+ 3(A1 B2 C2 + B1 C2 A2 + C1 A2 B2 )x y z ,2 2 2
we have
With regard to
∗∗∗
Based on Lemma above, Theorem 1 bellow gives for A ≤ 0 the necessary and
sufficient conditions to have f6 (x, y, z) ≥ 0 for all real numbers x, y, z which satisfy
k1 p2 + k2 q ≥ 0, (3.4)
where k1 and k2 are given real numbers (see P 2.75 and Remark 3 in Volume 1,
page 112).
Theorem 1 (Vasile Cirtoaje, 2008). Let f6 (x, y, z) be a symmetric homogeneous
polynomial of degree six which has the highest coefficient A ≤ 0. The inequality
f6 (x, y, z) ≥ 0 holds for all real numbers x, y, z satisfying (3.4) if and only if
f6 (x, 1, 1) ≥ 0
for all real x satisfying k1 (x + 2)2 + k2 (2x + 1) ≥ 0.
For k1 = k2 = 0, we get
Corollary 1. Let f6 (x, y, z) be a symmetric homogeneous polynomial of degree six
which has the highest coefficient A ≤ 0. The inequality f6 (x, y, z) ≥ 0 holds for all
real numbers x, y, z if and only if
f6 (x, 1, 1) ≥ 0
for all real x.
188 Vasile Cîrtoaje
∗∗∗
f6 (x, y, z) ≥ 0,
ξ ∈ (−∞, 0) ∪ (3, ∞)
• In this chapter, we consider that the function f¯6 (x, y, z) depends on only two
parameters α and β. Let us define the following nonnegative functions:
2
hα,β (x) = x + α(x + 2)(2x + 1) + β(x + 2)3 , (3.8)
Highest Coefficient Cancellation Method for Real Variables 189
where α and β are real numbers. Note that the parameters α and β of the function
fα,β may be also conjugate complex numbers.
• The function fα,β (x) has been derived by setting y = z = 1 in the associated
function 2
b1 2 c1 q2
f¯6 (x, y, z) = r − pq + p3 + · (3.9)
a1 a1 a1 p
with
which satisfies
therefore,
2
b1 2 c1 (2x + 1)2
fα,β (x) = f¯6 (x, 1, 1) = x − (x + 2)(2x + 1) + (x + 2)3 + · .
a1 a1 a1 x +2
Setting
β = −2, β = −1, β = 0, β = 1, β → ∞
in (3.6), we get in succession:
4(x − 1)6
f1,∞ (x) = ,
81(x + 2)2
(x − 1)4
f∞,∞ (x) = .
9(x + 2)2
Notice that the relative degree of the rational functions fα,β (x), fα,∞ (x) and f∞,∞ (x)
are six, four and two, respectively.
which satisfies
f¯6 (α, 1, 1) = 0.
Therefore,
2
α + (α + 2)(2α + 1)β
gα,β (x) = f¯6 (x, 1, 1) = x + β(x + 2)(2x + 1) − (x + 2)3 .
(α + 2)3
Setting
α = 0, α = −1, α = 1, β =0
in (3.7), we get:
2
β
ḡ0,β (x) = (x + 2)(x − 4) − 1 ,
4
2
ḡ−1,β (x) = x 2 + 5x + 8 + β(x + 2)(x + 5) ,
2
1 x +8
ḡ1,β (x) = (x − 1)2 + β(x + 2) ,
9 3
[αx 2 + α(α + 6)x − 8]2
ḡα,0 (x) = ,
(α + 2)6
In particular,
ḡ0,0 (x) = 1,
ḡ−1,0 (x) = (x 2 + 5x + 8)2 ,
(x − 1)2 (x + 8)2
ḡ1,0 (x) = .
729
Therefore,
2
hα,β (x) = f¯6 (x, 1, 1) = x + α(x + 2)(2x + 1) + β(x + 2)3 .
• Let
ξ ∈ (−∞, 0) ∪ (3, ∞).
For y = z = 1, the condition
p2 < ξq
involves
x ∈ , = ξ − 2 − ξ2 − 3ξ , ξ − 2 + ξ2 − 3ξ , (3.10)
while the condition
p2 ≥ ξq
involves
x ∈ \ .
Using the substitution
(2 + η)2
ξ= ,
1 + 2η
we have
4−η −1
= η, , η ∈ (−∞, −2) ∪ ,1 ,
1 + 2η 2
4−η −1
= ,η , η ∈ −2, ∪ (1, ∞).
1 + 2η 2
Thus, we can check that
and
η=1 ⇒ ξ = 3, = (1, 1) = ,
η=2 ⇒ ξ = 16/5, = (2/5, 2),
η=3 ⇒ ξ = 25/7, = (1/7, 3),
η=4 ⇒ ξ = 4, = (0, 4),
···
η→∞ ⇒ ξ → ∞, = (−1/2, ∞).
192 Vasile Cîrtoaje
Proof. Let
2
q2
E1 (x, y, z) = r + A1 pq + A2 p3 + A3
p
and 2
q2
F1 (x, y, z) = r + B1 pq + B2 p3 + B3
p
be the functions associated to Eα,β (x) and Fγ,δ (x), respectively; this means that
If
f6 (0, 1, −1) = 0,
then
f (−2) = 0.
Proof. Write f6 (x, y, z) in the form
where
g1 (p, q) = Bp3 + C pq,
g2 (p, q) = Dp6 + E p4 q + F p2 q2 + Gq3 ,
For (x, y, z) = (0, 1, −1), we have r = 0, p = 0 and q = −1, therefore
f6 (−2, 1, 1) = 4A.
Since
fα,−2 (−2) = 4, gα,β (−2) = 4, hα,β (−2) = 4,
we get
f (−2) = f6 (−2, 1, 1) − AEα,β (−2) = 4A − 4A = 0.
194 Vasile Cîrtoaje
Remark 1. The function f¯6 (x, y, z) associated to fα,β (x) (given by (3.9)) is zero for
If β = −2, then the function associated to fα,−2 (x) has the expression
2
α+8 2
f¯6 (x, y, z) = r − pq + p3 , (3.11)
9α + 18 9α + 18
and is zero for
In addition, if α → ∞, then
2
1
f¯6 (x, y, z) = r − pq
9
is zero for
If β → ∞, then
2
1 2α + 4 q2
f¯6 (x, y, z) = r + pq − ·
6α + 3 6α + 3 p
is zero for
is zero for
(x, y, z) = (1, 1, 1), (x, y, z) = (1, 0, 0).
Remark 2. If
f6 (x, 1, 1) = Afγ,−2 (x)
for all x ∈ , then there is k ≥ 0 such that the following identity holds:
is the best possible in the inequality f6 (x, y, z) ≥ 0 , then k = 0 and the following
identity holds:
2
γ+8 2
f6 (x, y, z) = A r − pq + p3 . (3.12)
9γ + 18 9γ + 18
Remark 3. Theorem 2 is also valid for the case where the parameters α and β of
the function fα,β (x) are conjugate complex numbers. For example, if k > 0 and
α= −k, β =− −k,
(x − 1)4 (x 2 + k)2
f
−k,−
−k = . (3.13)
9(k − 2)2 (x + 2)2
f6 (0, 1, −1) = 0.
• If
f6 (x, 1, 1) = (x − α)2 g(x), α = −2,
where g is a polynomial function, then the condition (a) in Theorem 2 applied for
Eα,β = gα,β , −2 ∈ ,
ḡ(−2) = 0.
196 Vasile Cîrtoaje
α (α + 2)2 g (−2)
β= + . (3.15)
3(α + 2) 12A
with δ given by
γ (γ + 2)2 g (−2)
δ= + (3.16)
3(γ + 2) 12A
and
2
γx 2 + γ(γ + 6)x − 8 δ(x + 2)(2γx + x + γ − 4)
ḡγ,δ (x) = + . (3.17)
(γ + 2)3 (γ + 2)2
Eα,β = hα,β , −2 ∈ ,
f (−2) = 0.
1 h (−2)
α= + , (3.18)
3 12A
where
h(x) = f6 (x, 1, 1).
Highest Coefficient Cancellation Method for Real Variables 197
with
1 h (−2)
γ= + (3.19)
3 12A
and 2
hγ,δ (x) = x + γ(x + 2)(2x + 1) + δ(x + 2)3 . (3.20)
k1 (x + y + z)2 + k2 (x y + yz + z x) ≥ 0,
where k1 and k2 are two fixed real numbers. More precisely, all conditions con-
cerning f6 (x, 1, 1) in Theorem 2 and Corollary 2 need to be satisfied only for
k1 (x + 2)2 + k2 (2x + 1) ≥ 0.
p = x + y + z, q = x y + yz + z x, r = x yz,
in the form
g(r) = A(p, q)r 2 + g3 (p, q)r + g4 (p, q),
where A(p, q), g3 (p, q) and g4 (p, q) are polynomial functions. The highest polyno-
mial A(p, q) of f8 (x, y, z) has the form
A(p, q) = μ1 p2 + μ2 q, (3.21)
∗∗∗
f8 (x, 1, 1) ≥ 0
∗∗∗
k1 (x + y + z)2 + k2 (x y + yz + z x) ≥ 0,
where k1 and k2 are two fixed real numbers. More precisely, all conditions con-
cerning f8 (x, 1, 1) need to be satisfied for
k1 (x + 2)2 + k2 (2x + 1) ≥ 0.
200 Vasile Cîrtoaje
Highest Coefficient Cancellation Method for Real Variables 201
3.2 Applications
2x 2 + 3 yz 2 y 2 + 3z x 2z 2 + 3x y 5
+ + ≤ .
4x 2 + y 2 + z 2 4 y 2 + z 2 + x 2 4z 2 + x 2 + y 2 2
9x 2 − 4 yz 9 y 2 − 4z x 9z 2 − 4x y 15
+ + ≥ .
3x 2 + 2 y 2 + 2z 2 3 y 2 + 2z 2 + 2x 2 3z 2 + 2x 2 + 2 y 2 7
x 2 + y 2 + z 2 ≥ 2(x y + yz + z x),
then
x 2 − 6 yz y 2 − 6z x z 2 − 6x y
+ 2 + 2 ≥ 0.
y2 + yz + z 2 z + zx + x 2 x + x y + y2
8x 2 + 3 yz 8 y 2 + 3z x 8z 2 + 3x y
+ 2 + 2 ≥ 11.
y + yz + z
2 2 z + zx + x 2 x + x y + y2
202 Vasile Cîrtoaje
8x 2 − 5 yz 8 y 2 − 5z x 8z 2 − 5x y
+ 2 + 2 ≥ 9.
y − yz + z
2 2 z − zx + x 2 x − x y + y2
5x 2 + 2 yz 5 y 2 + 2z x 5z 2 + 2x y
+ 2 + 2 ≥ 3.
2 y2 + 3 yz + 2z 2 2z + 3z x + 2x 2 2x + 3x y + 2 y 2
x( y + z) − yz y(z + x) − z x z(x + y) − x y 3
+ + ≤ .
x 2 + 3 y 2 + 3z 2 y 2 + 3z 2 + 3x 2 z 2 + 3x 2 + 3 y 2 7
−3
3.27. Let x, y, z be real numbers. If −2 ≤ k ≤ , then
2
k2
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) ≥ 1 + k − (x − y)2 ( y − z)2 (z − x)2 ,
4
3
(x 2 + y 2 )( y 2 + z 2 )(z 2 + x 2 ) ≥ 8x 2 y 2 z 2 + (x − y)2 ( y − z)2 (z − x)2 .
8
15
(x 2 +2 y 2 +2z 2 )( y 2 +2z 2 +2x 2 )(z 2 +2x 2 +2 y 2 ) ≥ 125x 2 y 2 z 2 + (x− y)2 ( y−z)2 (z−x)2 .
2
15
(2x 2 + y 2 +z 2 )(2 y 2 +z 2 + x 2 )(2z 2 + x 2 + y 2 ) ≥ 64x 2 y 2 z 2 + (x − y)2 ( y −z)2 (z− x)2 .
4
13 − 3 17 13 + 3 17
3.40. Let x, y, z be real numbers. If ≤k≤ , then
2 2
2
k
(x 2 + k yz)2 (x − y)(x − z) + − 1 (x − y)2 ( y − z)2 (z − x)2 ≥ 0.
4
k x 2 + 2 yz k y 2 + 2z x kz 2 + 2x y k−1
+ + ≥ .
k x 2 + y 2 + z 2 k y 2 + z 2 + x 2 kz 2 + x 2 + y 2 k+1
x 2 − yz y2 − z x z2 − x y
+ + 2 ≤ 1.
3x 2 + y +z
2 2 3y + z + x
2 2 2 3z + x 2 + y 2
x 2 + 3 yz y 2 + 3z x z 2 + 3x y
+ 2 + 2 ≥ 1.
y2 − yz + z 2 z − zx + x 2 x − x y + y2
2
x 6 + y 6 + z 6 − 3x 2 y 2 z 2 + (x 2 + k yz)( y 2 + kz x)(z 2 + k x y) ≥ 0.
k
13x − 1 13 y − 1 13z − 1 3
+ 2 + ≤ .
x 2 + 23 y + 23 z 2 + 23 2
x 2 + 2 yz y 2 + 2z x z 2 + 2x y 3(x y + yz + z x)
+ 2 + 2 ≥ .
y2 + yz + z 2 z + zx + x 2 x + x y + y2 x 2 + y 2 + z2
x 2 − 2 yz y 2 − 2z x z 2 − 2x y 3(x y + yz + z x)
+ 2 + 2 + ≥ 0.
y2 − yz + z 2 z − zx + x 2 x − x y + y2 x 2 + y 2 + z2
Highest Coefficient Cancellation Method for Real Variables 211
x2 y2 z2 (x + y + z)2
+ 2 + 2 ≥ 2 .
y2 − yz + z 2 z − zx + x 2 x −xy+ y 2 x + y 2 + z2
x 2 + 4 yz y 2 + 4z x z 2 + 4x y 15(x y + yz + z x)
(a) + 2 + 2 ≥ ;
y 2 + z2 z + x2 x + y2 2(x 2 + y 2 + z 2 )
2x 2 + 9 yz 2 y 2 + 9z x 2z 2 + 9x y 33(x y + yz + z x)
(b) + 2 + ≥ .
y 2 + z2 z + x2 x2 + y2 2(x 2 + y 2 + z 2 )
x2 y2 z2 4(x y + yz + z x)
+ + ≥ .
( y − z) 2 (z − x) 2 (x − y)2 x 2 + y 2 + z2
x2 y2 z2 (x + y + z)2
+ + ≥ 2 .
( y − z) 2 (z − x) 2 (x − y) 2 x + y 2 + z2
2x y 2 yz 2z x 4(x y + yz + z x)
+ + +1≥ .
x 2 + y 2 y 2 + z2 z2 + x 2 x 2 + y 2 + z2
212 Vasile Cîrtoaje
Highest Coefficient Cancellation Method for Real Variables 213
3.3 Solutions
Since
(x + 2 y)(x + 2z) = 2q + x 2 + 2 yz,
(2x + y)(2x + z) = 2q + 4x 2 − yz,
(x − y)(x − z) = x 2 + 2 yz − q,
( y − z)2 = −x 2 − 2 yz + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as
where
P1 (x, y, z) = (x 2 + 2 yz)(4x 2 − yz)(x 2 + 2 yz),
(
P3 (x, y, z) = (−x 2 − 2 yz).
According to (3.1) and (3.3), we get
The equality holds for x = y = z, for −x/2 = y = z (or any cyclic permutation),
and for −2x = y = z (or any cyclic permutation).
A1 = P1 (1, 1, 1) = 81.
Since the product (x − y) ( y − z)2 (z − x)2 has the highest coefficient equal to −27,
2
A = A1 + 3(−27) = 0.
By Corollary 1, we only need to show that f6 (x, 1, 1) ≥ 0 for all real x. This is true
because
f (x, 1, 1) = (x 2 − 4)2 (x − 1)2 ≥ 0.
The equality holds for x = y = z, for x + y + z = 0, and for x/2 = y = z (or any
cyclic permutation).
Observation. The inequality is equivalent to
(x − y)(x − z) = x 2 + 2 yz − q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = (x 2 + yz)(x 2 + 2 yz)2 ,
A1 = P1 (1, 1, 1) = 54.
Since the product (x − y)2 ( y − z)2 (z − x)2 has the highest coefficient equal to −27,
f6 (x, y, z) has the highest coefficient
A = A1 + 2(−27) = 0.
According to Corollary 1, we only need to show that f6 (x, 1, 1) ≥ 0 for all real x.
Indeed,
f (x, 1, 1) = (x 2 + 1)(x − 1)2 (x + 2)2 ≥ 0.
2x 2 + 3 yz 2 y 2 + 3z x 2z 2 + 3x y 5
+ + 2 ≤ .
4x + y + z
2 2 2 4y + z + x
2 2 2 4z + x 2 + y 2 2
Since
4x 2 + y 2 + z 2 = 3x 2 + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as f (x, y, z), where
(
f (x, y, z) = 5 (3x 2 ) − 2 (2x 2 + 3 yz)(3 y 2 )(3z 2 )
= 135x 2 y 2 z 2 − 108x 2 y 2 z 2 − 54 y 3z3
= 27x y z − 54
2 2 2 3 3
y z ,
216 Vasile Cîrtoaje
that is
A = 27 − 162 = −135.
According to Corollary 1, we only need to prove the original inequality for
y = z = 1.
2x 2 + 3 2(3x + 2) 5
+ ≤ ,
2(2x 2 + 1) x2 + 5 2
which is equivalent to
(x − 1)2 (2x − 1)2 ≥ 0.
The equality holds for x = y = z and for 2x = y = z (or any cyclic permutation).
9x 2 − 4 yz 9 y 2 − 4z x 9z 2 − 4x y 15
+ + ≥ .
3x 2 + 2 y 2 + 2z 2 3 y 2 + 2z 2 + 2x 2 3z 2 + 2x 2 + 2 y 2 7
(Vasile C., 2012)
Since
3x 2 + 2 y 2 + 2z 2 = x 2 + 2(p2 − 2q),
f6 (x, y, z) has the same highest coefficient A as f (x, y, z), where
(
f (x, y, z) = 7 (9x 2 − 4 yz)( y 2 )(z 2 ) − 15 (x 2 )
= 174x 2 y 2 z 2 − 28 y 3z3,
that is
A = 174 − 84 = 90.
Since A > 0 and
3x
The equality holds for x = y = z, and also for = y = z (or any cyclic
4
permutation).
k(k−3) 2
k−1 x + 2 yz 3(k + 1)(k − 2)
≤ ,
k x 2 + y 2 + z2 (k − 1)(k + 2)
kx
with equality for x = y = z, and for = y = z (or any cyclic permutation).
2
For
(
f6 (x, y, z) = m (k x 2 + y 2 + z 2 ) − (nx 2 + 2 yz)(k y 2 + z 2 + x 2 )(kz 2 + x 2 + y 2 ),
where
3(k + 1)(k − 2) k(k − 3)
m= , n= ,
(k − 1)(k + 2) k−1
we have
9(k + 1)(k − 1)(2 − k)
A=
k+2
and
2
f6 (x, 1, 1) = (x 2 + k + 1)(x − 1)2 (k x − 2)2 .
k+2
For k ∈ (0, 1) ∪ [2, ∞), we have A ≤ 0. According to Corollary 1, we only need
to prove that f6 (x, 1, 1) ≥ 0 for x ∈ , which is clearly true.
For k ∈ (1, 2), we have A > 0. According to Corollary 2, it suffices to prove that
for x ∈ . We have
(x − 1)4 (k x − 2)2
f2/k,−2 (x) = ,
81(k + 1)2
(k − 1)(2 − k)(x − 1)4 (k x − 2)2
Af2/k,−2 (x) = ,
9(k + 1)(k + 2)
(x − 1)2 (k x − 2)2 f (x)
f6 (x, 1, 1) − Af2/k,−2 (x) = ,
9(k + 1)(k + 2)
where
f (x) = 18(k + 1)(x 2 + k + 1) − (k − 1)(2 − k)(x − 1)2 .
Since
2(x 2 + k + 1) − (x − 1)2 = (x + 1)2 + 2k > 0,
we have
f (x) > 9(k + 1)(x − 1)2 − (k − 1)(2 − k)(x − 1)2 = (k2 + 6k + 11)(x − 1)2 ≥ 0.
x 2 + y 2 + z 2 ≥ 2(x y + yz + z x),
then
x 2 − 6 yz y 2 − 6z x z 2 − 6x y
+ 2 + 2 ≥ 0.
y2+ yz + z 2 z + zx + x 2 x + x y + y2
(Vasile C., 2012)
Since
x 2 + x y + y 2 = p2 − 2q − z 2 + x y, x 2 + xz + z 2 = p2 − 2q − y 2 + xz,
We have
x 3 + x 2 − 5x − 42 ≥ 5x 2 − 5x − 42 ≥ 15x − 42 > 0.
8x 2 + 3 yz 8 y 2 + 3z x 8z 2 + 3x y
+ 2 + 2 ≥ 11.
y + yz + z
2 2 z + zx + x 2 x + x y + y2
Since
y 2 + yz + z 2 = yz − x 2 + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as
where
(
P2 (x, y, z) = (8x 2 + 3 yz)(x y − z 2 )(xz − y 2 ), P3 (x, y, z) = ( yz − x 2 ).
Therefore,
A = P2 (1, 1, 1) − 11P3 (1, 1, 1) = 0.
According to Corollary 1, we only need to prove the original inequality for y = z =
1. Thus, we need to show that
8x 2 + 3 2(8 + 3x)
+ 2 ≥ 11,
3 x + x +1
220 Vasile Cîrtoaje
which is equivalent to
4x 4 + 4x 3 − 11x 2 − 6x + 9 ≥ 0,
8x 2 − 5 yz 8 y 2 − 5z x 8z 2 − 5x y
+ + ≥ 9.
y 2 − yz + z 2 z 2 − z x + x 2 x 2 − x y + y 2
(Vasile C., 2012)
From
f6 (x, y, z) = (8x 2 − 5 yz)(p2 − 2q − z 2 − x y)(p2 − 2q − y 2 − xz)
(
−9 (p2 − 2q − x 2 − yz),
where
(
P2 (x, y, z) = (8x 2 − 5 yz)(z 2 + x y)( y 2 + xz), P3 (x, y, z) = (x 2 + yz),
Since
The equality holds for x = y = z, and for −2x = y = z (or any cyclic permuta-
tion).
−2x
with equality for x = y = z, and for = y = z (or any cyclic permutation).
k+2
5x 2 + 2 yz 5 y 2 + 2z x 5z 2 + 2x y
+ 2 + 2 ≥ 3.
2 y2 + 3 yz + 2z 2 2z + 3z x + 2x 2 2x + 3x y + 2 y 2
From
f6 (x, y, z) = (5x 2 + 2 yz)(3x y − 2z 2 + 2p2 − 4q)(3xz − 2 y 2 + 2p2 − 4q)
(
−3 (3 yz − 2x 2 + 2p2 − 4q),
where
(
P2 (x, y, z) = (5x 2 +2 yz)(3x y −2z 2 )(3xz−2 y 2 ), P3 (x, y, z) = (3 yz−2x 2 ),
that is,
A = P2 (1, 1, 1) − 3P3 (1, 1, 1) = 21 − 3 = 18.
222 Vasile Cîrtoaje
Since
f6 (x, 1, 1)
= (5x 2 + 2)(2x 2 + 3x + 2) + 14(5 + 2x) − 21(2x 2 + 3x + 2)
2x 2 + 3x + 2
= 10x 4 + 15x 3 − 28x 2 − 29x + 32
= (x − 1)2 (10x 2 + 35x + 32).
Since the original inequality is an equality for (x, y, z) = (0, 1, −1), that is
f6 (0, 1, −1) = 0,
we have f (−2) = 0 for all real γ (see Proposition 1). To have f (x) ≥ 0 in the
vicinity of x = −2, the condition f (−2) = 0 is necessary. This condition involves
γ = −50/37 and
The equality holds for x = y = z, and for x = 0 and y + z = 0 (or any cyclic
permutation).
with equality for x = y = z, and also for x = 0 and y + z = 0 (or any cyclic
permutation).
Highest Coefficient Cancellation Method for Real Variables 223
Since
7x 2 + y 2 + z 2 = 6x 2 + p2 − 2q, (x + y)(x + z) = x 2 + q,
f6 (x, y, z) has the same highest coefficient A as
(
4 (6x 2 ) − 3 x 2 (6 y 2 )(6z 2 ),
that is
A = 4 · 63 − 9 · 62 = 540.
Since
f6 (x, 1, 1)
= 4(7x 2 + 2)(x 2 + 8) − 3(x + 1)[(x + 1)(x 2 + 8) + 4(7x 2 + 2)]
x2 + 8
= 25x 4 − 90x 3 + 121x 2 − 72x + 16 = (x − 1)2 (5x − 4)2 ,
f (x) = 147(x 2 + 8) − 20(x − 1)2 > 40(x 2 + 8) − 20(x − 1)2 = 20[(x + 1)2 + 14] > 0.
Thus, the proof is completed. The equality holds for x = y = z, and also for
x y z
= = (or any cyclic permutation).
4 5 5
224 Vasile Cîrtoaje
Since
that is
A = −24 + 168 = 144.
Since
f6 (x, 1, 1)
= 3(x 2 + 6)(3x 2 + 4) − 7[(2x − 1)(3x 2 + 4) + 2(x 2 + 6)]
3x 2 + 4
= 9x 4 − 42x 3 + 73x 2 − 56x + 16 = (x − 1)2 (3x − 4)2 ,
Thus, the proof is completed. The equality holds for x = y = z, and also for
x y z
= = (or any cyclic permutation).
4 3 3
226 Vasile Cîrtoaje
A1 = P1 (1, 1, 1) = 54.
A = A1 + (−27) = 27.
Since
f6 (x, 1, 1) = f (x, 1, 1) = 2(2x 2 + 1)(x − 1)2 ,
we apply Corollary 2 for
4(x − 1)4
Fγ,δ (x) = f∞,−2 (x) = .
81
We need to show that
f6 (x, 1, 1) ≥ Af∞,−2 (x)
for x ∈ . Indeed, we get
Since
(x − y)(x − z) = x 2 + 2 yz − q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = (x 2 − yz)2 (x 2 + 2 yz),
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, y = 1 and z = −1 in the inequality
(x 2 − yz)2 (x − y)(x − z) ≥ k(x − y)2 ( y − z)2 (z − x)2 ,
3
we get k ≤ . According to (3.12) from Remark 2, the identity holds:
4
1
f6 (x, y, z) = (9r − 7pq + 2p3 )2 .
4
Therefore, the original inequality is equivalent to
7k2 − 20k − 20
f6 (x, y, z) = f (x, y, z) + (x − y)2 ( y − z)2 (z − x)2 ,
24
f (x, y, z) = (x 2 + yz)(x − y)(x − z)(x − k y)(x − kz).
Since
(x − y)(x − z) = x 2 + 2 yz − q,
(x − k y)(x − kz) = x 2 + (k + k2 ) yz − kq,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = (x 2 + yz)(x 2 + 2 yz)[x 2 + (k + k2 ) yz];
A1 = P1 (1, 1, 1) = 18(1 + k + k2 ).
For k = −2, we have A = 0. Since f6 (x, 1, 1) ≥ 0 for any real x, the conclusion
follows by Corollary 1.
For k = −2, we apply Corollary 2 for
yz zx xy 1
+ + + ≥ 0.
( y − z)2 (z − x)2 (x − y)2 4
1
f6 (x, y, z) = f (x, y, z) + (x − y)2 ( y − z)2 (z − x)2 ,
4
f (x, y, z) = yz(x − y)2 (x − z)2 .
Since
(x − y)(x − z) = x 2 + 2 yz − q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = yz(x 2 + 2 yz)2 ;
A1 = P1 (1, 1, 1) = 27.
27 81
A = A1 − = .
4 4
Since
f6 (x, 1, 1) = (x − 1)4 ,
we apply Corollary 2 for
4(x − 1)4
Fγ,δ (x) = f∞,−2 (x) = .
81
We have
f6 (x, 1, 1) − Af∞,−2 (x) = 0.
Thus, the proof is completed. The inequality is an equality for all distinct real x, y, z
which satisfy
9x yz = (x + y + z)(x y + yz + z x).
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 in the in-
equality f6 (x, y, z) ≥ 0 is the best possible. Thus, setting x = 0, y = 1 and z = −1
in the inequality
yz(x − y)2 (x − z)2 + k(x − y)2 ( y − z)2 (z − x)2 ≥ 0,
Highest Coefficient Cancellation Method for Real Variables 231
1
we get k ≥ . According to (3.12) from Remark 2, the identity holds:
4
1
f6 (x, y, z) = (9r − pq)2 .
4
Therefore, for distinct x, y, z, the original inequality is equivalent to
(9r − pq)2 ≥ 0.
A1 = P1 (1, 1, 1) = 27(1 + k + k2 ).
For k = −2, we have A = 0. Since f6 (x, 1, 1) ≥ 0 for any real x, the conclusion
follows (by Corollary 1).
232 Vasile Cîrtoaje
x2 y2 z2
+ + ≥ 2,
( y − z)2 (z − x)2 (x − y)2
Solution. Let
f (x, y, z) = (x 2 + 2 yz)(x 2 − y 2 )(x 2 − z 2 )
and
1
f6 (x, y, z) = f (x, y, z) + (x − y)2 ( y − z)2 (z − x)2 .
2
Since
(x − y)(x − z) = x 2 + 2 yz − q, (x + y)(x + z) = x 2 + q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = (x 2 + 2 yz)2 x 2 ,
Highest Coefficient Cancellation Method for Real Variables 233
Since
x 2 + y 2 + z 2 − x y − yz − z x = p2 − 3q,
f6 (x, y, z) has the highest coefficient
A = 81(k + 2)2 .
which is an identity.
The equality holds for all real x, y, z which satisfy
Observation 1. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, y = 1 and z = −1 in the inequality
(x − y)2 ( y − z)2 (z − x)2
(x − y)(x − z)(x − k y)(x − kz) ≥ αk ,
x2 + y 2 + z 2 − x y − yz − z x
3(k + 2)2
we get αk ≤ . According to (3.12) from Remark 2, the identity holds:
4
2
f6 (x, y, z) = 9(k + 2)r − (k + 8)pq + 2p3 .
Therefore, if x, y, z are not all equal, then the original inequality is equivalent to
2
9(k + 2)r − (k + 8)pq + 2p3 ≥ 0.
Highest Coefficient Cancellation Method for Real Variables 235
which is equivalent to
(x + y + z)2 (x 2 + y 2 + z 2 − x y − yz − z x) ≥ 0.
Solution. Let
f (x, y, z) = (x 2 + yz)(x − y)(x − z)(x − k y)(x − kz)
and
k2
f6 (x, y, z) = f (x, y, z) − 1 + k − (x − y)2 ( y − z)2 (z − x)2 .
4
Since
For k = −2, we only need to show that f6 (x, 1, 1) ≥ 0 for 2x +1 ≥ 0 (see Theorem
1). Indeed,
f6 (x, 1, 1) = (x 2 + 1)(x − 1)2 (x + 2)2 ≥ 0.
Consider next k = −2. Since A > 0 and
The equality holds for x = y = z, and for x/k = y = z (or any cyclic permuta-
tion) if k = 0 and 2k + 1 ≥ 0. If k = 0, then the equality holds also for x = 0 and
y = z (or any cyclic permutation).
Solution. Let
f6 (x, y, z) = (x 2 + 2 yz)2 (x − y)(x − z).
Since (x − y)(x − z) = x 2 + 2 yz − q, f6 (x, y, z) has the same highest coefficient as
P1 (x, y, z) = (x 2 + 2 yz)3 ,
where
Choosing γ = 1, we get
g1 (x) = (x + 2)2 ≥ 0,
and
we select
Fγ,δ = g1,δ
with δ given by (3.16). We have
where 2
1 x +8 4
ḡ1,δ (x) = (x − 1)2 + δ(x + 2) = (x − 1)4 .
9 9 729
We have
4
g(x) − Aḡγ,δ (x) = (x 2 + 2)2 − (x − 1)4
9
1
= (x + 2)2 (5x 2 − 4x + 8) ≥ 0.
9
Highest Coefficient Cancellation Method for Real Variables 239
Solution. Let
f (x, y, z) = x 2 (x 2 + yz)(x − y)(x − z)
and
f6 (x, y, z) = f (x, y, z) − (x − y)2 ( y − z)2 (z − x)2 .
Since
(x − y)(x − z) = x 2 + 2 yz − q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = x 2 (x 2 + yz)(x 2 + 2 yz),
A = 18 − (−27) = 45.
Since
f6 (x, 1, 1) = x 2 (x 2 + 1)(x − 1)2 ,
x 2 (x − 1)4
f0,−2 (x) = ,
81
5x 2 (x − 1)4
f6 (x, 1, 1) − Af0,−2 (x) = x 2 (x 2 + 1)(x − 1)2 −
9
2x 2 (x − 1)2 (x + 2)(2x + 1)
= ,
9
we will apply Theorem 2 for
According to Proposition 1, the difference f6 (x, 1, 1) − Afγ,−2 (x) is zero for x = −2.
Therefore, we will use the Cauchy-Schwarz inequality
f6 (x, 1, 1) ≥ F (x),
1 2
F (x) = x (2x − 1)2 (x − 1)2 .
5
Thus, the condition (b) in Theorem 2 is satisfied if
where
10(x + 2)2
g1 (x) = ≥ 0,
17
and
10(35x 2 − 22x − 4) 10(−8x − 4) −40(2x + 1)
g2 (x) = > = ≥ 0.
17 17 17
Highest Coefficient Cancellation Method for Real Variables 241
The equality holds for x = y = z, for x = 0 and y = z (or any cyclic permutation),
and for x = 0 and y + z = 0 (or any cyclic permutation).
Observation 1. Similarly, applying Theorem 2 for
13k − 4
ξ → ∞, Eα,β = f k,−2 , Fγ,δ = fγ,−2 , γ= ,
k + 17
we can prove the following generalization:
−7
• Let x, y, z be real numbers. If ≤ k ≤ 1, then
17
k2
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) ≥ 1 + k − (x − y)2 ( y − z)2 (z − x)2 ,
4
with equality for x = y = z, for x/k = y = z (or any cyclic permutation) if k = 0,
and for x = 0 and y + z = 0 (or any cyclic permutation). If k = 0, then the equality
holds also for x = 0 and y = z (or any cyclic permutation).
We have
f6 (x, y, z) = (x 2 + yz)(x − y)(x − z)(x − k y)(x − kz)
k2
− 1+k− (x − y)2 ( y − z)2 (z − x)2 ,
4
45(k + 2)2
A= ,
4
−1 −1
= , ∞ , \ = −∞, ,
2 2
f (x, 1, 1) = (x 2 + 1)(x − 1)2 (x − k)2 ,
4(x − 1)4 (x − k)2 5(x − 1)4 (x − k)2
f k,−2 (x) = , Af k,−2 (x) = ,
81(2 + k)2 9
2(x − 1)2 (x − k)2 (x + 2)(2x + 1)
f6 (x, 1, 1) − Af k,−2 (x) = ,
9
1
F (x) = (2x − 1)2 (x − 1)2 (x − k)2 ,
5
4(x − 1)4 [(k + 17)x − 13k + 4]2
fγ,−2 (x) = ,
81 · 225(k + 2)2
(x − 1)4 [(k + 17)x − 13k + 4]2
Afγ,−2 (x) = ,
405
(x − 1)2 g1 (x)g2 (x)
F (x) − Afγ,−2 (x) = ,
405
g1 (x) = (1 − k)(x + 2)2 ≥ 0,
242 Vasile Cîrtoaje
1
x 2 + 1 ≥ (2x − 1)2 ,
5
we can prove the inequality from Observation 1 for the extended range
−8
≤ k ≤ 9 5 − 17 ≈ 3.1246.
19
We have
(x − 1)2 g1 (x)
f6 (x, 1, 1) − Afγ,−2 (x) = ,
405
Observation 3. Actually, the inequality from Observation 1 holds for k ∈ [−2, 28].
45(k + 2)2
A= .
4
Since
f (x, 1, 1) = (x 2 + 1)(x − 1)2 (x − k)2 ,
Highest Coefficient Cancellation Method for Real Variables 243
The condition (a) in Theorem 2, namely f6 (x, 1, 1) − Af k,−2 (x) ≥ 0 for x > −1/2,
is satisfied.
To prove the condition (b), namely f6 (x, 1, 1) ≥ Ahγ,δ (x) for x ≤ −1/2, we see
−1
that −2 ∈ −∞, and
2
k2
f6 (0, y, z) = k2 y 3 z 3 +( y −z) ( y 5 − z 5 ) − k yz( y 3 − z 3 ) − 1 + k − y 2 z 2 ( y −z)2 ,
4
1 h (−2)
γ= + ,
3 12A
where
h(x) = f6 (x, 1, 1) = (x 2 + 1)(x − 1)2 (x − k)2 .
We have
h (x) = 2x(x − 1)2 (x − k)2 + 2(x 2 + 1)(x − 1)(x − k)2 + 2(x 2 + 1)(x − 1)2 (x − k),
Solution. This is the inequality of the preceding P 3.24 for k = 28. Having in
view the proof of P 3.24, we only need to show that there exists a real δ such that
f (x) ≥ 0 for x ≤ −1/2, where
45(k + 2)2
f (x) = (x 2 + 1)(x − 1)2 (x − k)2 − hγ,δ (x),
4
= (x 2 + 1)(x − 1)2 (x − 28)2 − 10125hγ,δ (x),
−7k − 44 −8
γ= = ,
45(k + 2) 45
2
hγ,δ (x) = x + γ(x + 2)(2x + 1) + δ(x + 2)3
2
8
= x− (x + 2)(2x + 1) + δ(x + 2)3 .
45
Actually, there is a unique δ such that f (x) ≥ 0 for all x ≤ −1/2, namely
17
δ= .
2250
For this value of δ, we have
The equality holds for x = y = z, for x/28 = y = z (or any cyclic permutation),
and for x = 0 and y + z = 0 (or any cyclic permutation).
Observation 1. Similarly, using
−7k − 44
γ=
45(k + 2)
and
17
δ= ,
75(k + 2)
therefore
2
7k + 44 17
hγ,δ (x) = x − (x + 2)(2x + 1) + (x + 2)3 ,
45(k + 2) 75(k + 2)
246 Vasile Cîrtoaje
Solution. Let
f (x, y, z) = (x 2 + yz)(x 2 − y 2 )(x 2 − z 2 )
and
1
f6 (x, y, z) = f (x, y, z) + (x − y)2 ( y − z)2 (z − x)2 .
4
Since
(x − y)(x − z) = x 2 + 2 yz − q, (x + y)(x + z) = x 2 + q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = (x 2 + yz)(x 2 + 2 yz)x 2 ,
1 45
A = 18 + (−27) = .
4 4
Since
f6 (x, 1, 1) = (x 2 + 1)(x 2 − 1)2 ,
4(x − 1)4 (x + 1)2
f−1,−2 (x) = ,
81
Highest Coefficient Cancellation Method for Real Variables 247
The condition (a) in Theorem 2, namely f6 (x, 1, 1)−Af−1,−2 (x) ≥ 0 for x > −1/2,
is satisfied.
To provethe condition (b), namely f6 (x, 1, 1) ≥ Ag−1,δ (x) for x ≤ −1/2, we see
−1
that −2 ∈ −∞, and
2
1 2 2
f6 (0, y, z) = y 3 z 3 + ( y 2 − z 2 )( y 4 − z 4 ) + y z ( y − z)2 ,
4
f6 (0, 1, −1) = −1 + 0 + 1 = 0.
According to Proposition 1, the difference f6 (x, 1, 1) − Ag−1,δ (x) is zero for x = −2.
Therefore, we will use the inequality
1
x 2 + 1 ≥ (2x − 1)2 ,
5
which is an equality for x = −2, to get
f6 (x, 1, 1) ≥ F (x),
1
F (x) = (2x − 1)2 (x 2 − 1)2 .
5
Thus, the condition (b) in Theorem 2 is satisfied if
where
(2x − 1)2 (x − 1)2 45 2
g(x) = − [x + 5x + 8 + δ(x + 2)(x + 5)]2
5 4
g1 (x)g2 (x)
= ,
20
4(x + 2)2
g1 (x) = ≥ 0,
3
and
4(5x 2 − 13x − 1)
g2 (x) = > 0.
3
The equality holds for x = y = z, for −x = y = z (or any cyclic permutation),
and for x = 0 and y + z = 0 (or any cyclic permutation).
Observation 1. Similarly, applying Theorem 2 for
−7k − 44
ξ → ∞, Eα,β = f k,−2 , Fγ,δ = g k,δ , δ= ,
45(k + 2)
we can prove the following generalization:
−5
• Let x, y, z be real numbers. If ≤ k ≤ 1, then
3
k2
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) ≥ 1 + k − (x − y)2 ( y − z)2 (z − x)2 ,
4
with equality for x = y = z, for x/k = y = z (or any cyclic permutation) if k = 0,
and for x = 0 and y + z = 0 (or any cyclic permutation). If k = 0, then the equality
holds also for x = 0 and y = z (or any cyclic permutation).
We have
f6 (x, y, z) = (x 2 + yz)(x − y)(x − z)(x − k y)(x − kz)
k2
− 1+k− (x − y)2 ( y − z)2 (z − x)2 ,
4
Highest Coefficient Cancellation Method for Real Variables 249
45(k + 2)2
A= ,
4
−1 −1
= , ∞ , \ = −∞, ,
2 2
f (x, 1, 1) = (x 2 + 1)(x − 1)2 (x − k)2 ,
4(x − 1)4 (x − k)2 5(x − 1)4 (x − k)2
f k,−2 (x) = , Af k,−2 (x) = ,
81(2 + k)2 9
2(x − 1)2 (x − k)2 (x + 2)(2x + 1)
f6 (x, 1, 1) − Af k,−2 (x) = ,
9
1
F (x) = (2x − 1)2 (x − 1)2 (x − k)2 ,
5
2 2
k x + k(k + 6)x − 8 δ(x + 2)(2k x + x + k − 4)
g k,δ (x) = (x − k)2 + ,
(k + 2)3 (k + 2)2
(x − k)2 g 2 (x)
Ag k,δ (x) = ,
180(k + 2)4
1
x 2 + 1 ≥ (2x − 1)2 ,
5
we can prove the inequality from Observation 1 for the extended range
−17 13 + 9 5
≤k≤ ≈ 8.281.
10 4
We have
(x − k)2 g1 (x)
f6 (x, 1, 1) − Ag k,γ (x) = ,
45(k + 2)2
250 Vasile Cîrtoaje
g1 (x) = 45(k + 2)2 (x 2 + 1)(x − 1)2 − [(7k + 11)x 2 − (5k + 22)x + 7k + 2]2
= (x + 2)2 g2 (x),
−3
P 3.27. Let x, y, z be real numbers. If −2 ≤ k ≤ , then
2
k2
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) ≥ 1 + k − (x − y)2 ( y − z)2 (z − x)2 ,
4
(Vasile C., 2014)
Solution. Let
f6 (x, y, z) = (x 2 + yz)(x − y)(x − z)(x − k y)(x − kz)
k2
− 1+k− (x − y)2 ( y − z)2 (z − x)2 .
4
As shown in the proof of P 3.21, f6 (x, y, z) has the highest coefficient
45(k + 2)2
A= .
4
In addition,
f6 (x, 1, 1) = (x 2 + 1)(x − 1)2 (x − k)2 .
For k = −2, we have A = 0. Since f6 (x, 1, 1) ≥ 0 for all real x, the conclusion
follows from Corollary 1. Consider further that
−3
−2 < k ≤ ,
2
and apply Theorem 2 for
−7k − 44
Eα,β = g k,β , β= , Fγ,δ = f k,∞ ,
45(k + 2)
4(k + 2)2
and η = 2k + 2 ∈ (−2, −1], which involves ξ = and
4k + 5
2 − 2k 2 − 2k
= , 2k + 2 , \ = −∞, ∪ [2k + 2, ∞).
5 + 4k 5 + 4k
Highest Coefficient Cancellation Method for Real Variables 251
for x ∈ . Since
1
f (x, 1, 1) ≥ F (x) = (2x − 1)2 (x − 1)2 (x − k)2 ,
5
it suffices to show that
F (x) ≥ Ag k,β (x).
As shown at Observation 1 of the preceding P 3.26, we have
g3 (x) = 13(k + 2)(2k + 3) + g4 (x) ≤ g4 (x), g4 (x) = (13k + 23)x − 3(11k + 24),
−x −3(11k + 24)x
−1 ≤ , −3(11k + 24) ≤ ,
2k + 2 2k + 2
we get
therefore g2 (2k + 2) ≥ 0.
Condition (b). We need to show that
for
2 − 2k
x ∈ \ = −∞, ∪ [2k + 2, ∞).
5 + 4k
Since
1
f (x, 1, 1) ≥ F (x) = (x − 1)4 (x − k)2 ,
2
it suffices to show that
F (x) ≥ Af k,∞ (x),
where
4(x − 1)4 (x − k)2
f k,∞ (x) = .
9(1 + 2k)2 (x + 2)2
We have
(x − 1)4 (x − k)2 g1 (x)
F (x) − Af k,∞ (x) = ,
2(1 + 2k)2 (x + 2)2
g1 (x) = (1 + 2k)2 (x + 2)2 − 10(k + 2)2 .
It suffices to show that g2 (x) ≥ 0, where
Indeed,
6(k + 2)
(1 + 2k)(x + 2) − 4(k + 2) ≥ (1 + 2k) − 4(k + 2)
5 + 4k
−2(k + 2)(2k + 7)
= > 0.
5 + 4k
Case 2: x ≥ 2k + 2. Since
x + 2 ≥ 2(k + 2) > 0,
the inequality g2 (x) ≥ 0 holds if
Indeed,
The equality holds for x = y = z, for x/k = y = z (or any cyclic permutation) if
k = 0, and for x = 0 and y + z = 0 (or any cyclic permutation). If k = 0, then the
equality holds also for x = 0 and y = z (or any cyclic permutation).
Observation 1. From P 3.23-3.27 and the observations attached to them, the fol-
lowing generalization follows:
• Let x, y, z be real numbers. If −2 ≤ k ≤ 28, then
k2
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) ≥ 1 + k − (x − y)2 ( y − z)2 (z − x)2 ,
4
with equality for x = y = z, for x/k = y = z (or any cyclic permutation) if k = 0,
and for x = 0 and y + z = 0 (or any cyclic permutation). If k = 0, then the equality
holds also for x = 0 and y = z (or any cyclic permutation).
Note that the coefficient of the product (x − y)2 ( y −z)2 (z− x)2 is the best possible.
Setting x = k, y = 1 + t and z = 1 − t, the inequality
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) ≥ δk (x − y)2 ( y − z)2 (z − x)2
turns into
A(k, δk )t 6 + B(k, δk )t 4 + C(k, δk )t 2 ≥ 0,
where
A(k, δk ) = −4δk + 4 + 4k − k2 .
From the necessary condition A(k, δk ) ≥ 0, we get
k2
δk ≤ 1 + k − .
4
then
(x 2 + yz)(x − y)(x − z)(x − k y)(x − kz) + δk (x − y)2 ( y − z)2 (z − x)2 ≥ 0.
Solution. Denote
f (x, y, z) = (x 2 + yz)(x − y)(x − z)(x − k y)(x − kz),
In addition,
f6 (x, 1, 1) = (x 2 + 1)(x − 1)2 (x − k)2 .
For k = −2, we have A = 0. Since f6 (x, 1, 1) ≥ 0 for any real x, the conclusion
follows from Corollary 1.
Consider further that −5 ≤ k < −2. Since
and
(k + 2)2
we apply Theorem 2 for η = k, which involves ξ = ,
2k + 1
4−k 4−k
= k, , \ = (−∞, k] ∪ ,∞ ,
2k + 1 2k + 1
and for
Eα,β = f k,−2 , Eγ,δ = f k,∞ .
Highest Coefficient Cancellation Method for Real Variables 255
The condition (a), namely f6 (x, 1, 1) ≥ Af k,−2 (x) for x ∈ , is satisfied since
4−k 3(k − 1)(k + 5)
(k − 4)x + (2k + 1) > (k − 4) · + (2k + 1) = ≥ 0,
2k + 1 2k + 1
4−k
(2k + 1)x + k − 4 > (2k + 1) · + k − 4 = 0.
2k + 1
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Af k,∞ (x) for x ∈ \ , where
f (x) = (k − 1)2 (2k + 1)2 (x 2 + 1)(x + 2)2 − (k + 2)2 (5k2 − 4k + 17)(x − 1)2 .
4−k
Case 2: x ≥ . Since
2k + 1
4−k
∈ (−2, −1],
2k + 1
4−k
h is decreasing on , 1 ∪ [3, ∞) and increasing on [1, 3]. Therefore, we
2k+ 1
4−k 4−k
have h(x) ≤ h if and only if h(3) ≤ h . On the other hand, since
2k + 1 2k + 1
4−k
h is decreasing on , −1 , we have
2k + 1
4−k
h ≥ h(−1).
2k + 1
Therefore, it suffices to show that h(3) ≤ h(−1). Since
2
h(−1) = 2, h(3) = ,
125
the conclusion follows.
The equality holds for x = y = z, and for x/k = y = z (or any cyclic permutation).
Observation 1. The coefficient δk of the product (x − y)2 ( y − z)2 (z − x)2 in the
original inequality is the best possible. Setting x = k, y = 1 + t and z = 1 − t, the
original inequality turns into
where
(k + 1)2 (k2 − 6k + 2)
δk∗ = .
4(k − 2)2
Highest Coefficient Cancellation Method for Real Variables 257
If −4 ≤ k ≤ −1, then
(x 2 + yz)(x − y)(x − z)(x − k y + z)(x − kz + y) + δk∗ (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
with equality for x = y = z, and for x/(k − 1) = y = z (or any cyclic permutation).
So, we only need to show that there exists a real number δ such that
Since
f (−5) = 0,
the condition f (−5) = 0 is necessary to have f (x) ≥ 0 in the vicinity of x = −5.
This condition involves
−59
δ=
81
258 Vasile Cîrtoaje
and
4g(x)
f (x) = ,
81
where
and
f6 (x, y, z) = f (x, y, z) − 7(x − y)2 ( y − z)2 (z − x)2 .
Since (x − y)(x − z) = x 2 + 2 yz − q and
x 2 + y 2 = p2 − 2q − z 2 , x 2 + z 2 = p2 − 2q − y 2 ,
A1 = P2 (1, 1, 1) = 36.
A = 36 − 7(−27) = 225.
Since
f6 (x, 1, 1) = (x − 1)2 (x 2 + 1)2 ,
we will apply Corollary 2 for
for x ∈ . We have
(x − 1)2 g(x)
f6 (x, 1, 1) − Afγ,−2 (x) = ,
9(2 + γ)2
where
The equality holds for x = y = z, and also for x = 0 and y + z = 0 (or any cyclic
permutation).
Solution. Let
f (x, y, z) = 4 (x − y)(x − z)(x 2 + y 2 )(x 2 + z 2 )
and
f6 (x, y, z) = f (x, y, z) − 15(x − y)2 ( y − z)2 (z − x)2 .
As shown at the preceding P 3.30, f (x, y, z) has the highest coefficient A1 = 36,
therefore f6 (x, y, z) has the highest coefficient
A = A1 − 15(−27) = 441.
Since
f6 (x, 1, 1) = 4(x − 1)2 (x 2 + 1)2 ,
we will apply Corollary 2 for
Having in view Remark 5, we need to show that there exists a real γ such that
for 2x + 1 ≥ 0. We have
where
g(x) = 9(2 + γ)2 (x 2 + 1)2 − 49(x − 1)2 (x − γ)2 .
Choosing γ = 1/3, which involves g(−1/2) = 0, we get
49
g(x) = 9(x 2 + 1)2 − (x − 1)2 (3x − 1)2
9
196(2x + 1)(3x 2 − 2x + 2)
= ≥ 0.
9
The equality holds for x = y = z, and also when x, y, z are proportional to the
roots of the equation
7t 3 − 21t 2 + 18 = 0.
Observation. The last equality condition follows from the necessary condition
f¯6 (x, y, z) = 0, where, according to (3.11),
2
γ+8 2 25 2 3 2
f¯6 (x, y, z) = r − pq + p3 = r − pq + p .
9γ + 18 9γ + 18 63 21
Solution. Let
f (x, y, z) = (x − y)(x − z)(x 2 + x y + y 2 )(x 2 + xz + z 2 )
and
3
f6 (x, y, z) = f (x, y, z) − (x − y)2 ( y − z)2 (z − x)2 .
4
Highest Coefficient Cancellation Method for Real Variables 261
x 2 + x y + y 2 = p2 − 2q − z 2 + x y, x 2 + xz + z 2 = p2 − 2q − y 2 + xz,
for x ∈ . We have
(x − 1)2 g(x)
f6 (x, 1, 1) − Afγ,−2 (x) = ,
(2 + γ)2
where
(x + 2)2 (5x 2 + 2x + 2)
g(x) = ≥ 0.
4
The equality holds for x = y = z, and also for x = 0 and y + z = 0 (or any cyclic
permutation).
Observation 1. Similarly, applying Corollary 2 for
1 − 4k
Fγ,δ = fγ,−2 , γ= ,
7−k
262 Vasile Cîrtoaje
Observation 2. Actually, the inequality from Observation 1 holds for k ∈ [−2, 5/2].
For k = 5/2 and k = 2, we get the inequalities in P 3.1 and P 3.47, respectively.
Solution. Let
f (x, y, z) = (x − y)(x − z)(x 2 + x y + y 2 )(x 2 + xz + z 2 )
and
f6 (x, y, z) = f (x, y, z) − 3(x − y)2 ( y − z)2 (z − x)2 .
Highest Coefficient Cancellation Method for Real Variables 263
A = 0 − 3(−27) = 81.
(x − 1)4 x 2
Fγ,δ (x) = f0,−2 (x) = .
81
Having in view Remark 5, we need to show that
The equality holds for x = y = z, and also when x, y, z are proportional to the
roots of the equation
t 3 − 3t 2 + 3 = 0.
Observation 1. The last equality condition follows from the necessary condition
f¯6 (x, y, z) = 0, where
2 2
γ+8 2 4 1
f¯6 (x, y, z) = r − pq + p3 = r − pq + p3
9γ + 18 9γ + 18 9 9
(see (3.11) from Remark 1). Moreover, if p = 3 and q = 0, then the condition
f¯6 (x, y, z) = 0 involves r = −3.
Observation 2. Similarly, applying Corollary 2 for
1−k
Fγ,δ = fγ,−2 , γ= ,
3
we can prove the following generalization:
• Let x, y, z be real numbers such that x y + yz + z x ≥ 0. If k ∈ [−2, 5/2], then
(3 − k)(5 + k)
(x− y)(x−z)(x 2 +k x y+ y 2 )(x 2 +k xz+z 2 ) ≥ (x− y)2 ( y−z)2 (z−x)2 ,
4
with equality for x = y = z, and also when x, y, z are proportional to the roots of the
equation
(7 − k)t 3 − 3(7 − k)t 2 + 18 = 0.
264 Vasile Cîrtoaje
For
f6 (x, y, z) = (x − y)(x − z)(x 2 + k x y + y 2 )(x 2 + k xz + z 2 )
(3 − k)(5 + k)
− (x − y)2 ( y − z)2 (z − x)2 ,
4
we have
9
A = (k − 7)2 ,
4
f6 (x, 1, 1) = (x − 1)2 (x 2 + k x + 1)2 ,
and, for 2x + 1 ≥ 0,
1
f6 (x, 1, 1) − Afγ,−2 (x) = (k + 2)(x − 1)2 (2x + 1)[6x 2 + 4(k − 1)x + 4 − k] ≥ 0.
9
3
(x 2 + y 2 )( y 2 + z 2 )(z 2 + x 2 ) ≥ 8x 2 y 2 z 2 + (x − y)2 ( y − z)2 (z − x)2 .
8
(Vasile C., 2011)
3
−x 2 y 2 z 2 − 8x 2 y 2 z 2 − (x − y)2 ( y − z)2 (z − x)2 ,
8
that is
81 9
A = −9 + = .
8 8
Since
f6 (x, 1, 1) = 2(x 2 + 1)2 − 8x 2 = 2(x 2 − 1)2 ,
4(x − 1)4 (x + 1)2
f−1,∞ (x) = ,
9(x + 2)2
3(x 2 − 1)2 (x + 1)(x + 5)
f6 (x, 1, 1) − Af−1,∞ (x) = ,
2(x + 2)2
Highest Coefficient Cancellation Method for Real Variables 265
The condition (b), namely f6 (x, 1, 1) − Af−1,∞ (x) ≥ 0 for x ∈ (−∞, −5] ∪
[−1, ∞), is satisfied.
The condition (a) is satisfied if f6 (x, 1, 1) − Af−1,−2 (x) ≥ 0 for x ∈ (−5, −1). We
have
4(x − 1)4 (x + 1)2
f−1,−2 (x) = ,
81
(x 2 − 1)2 (x + 5)(7 − x)
f6 (x, 1, 1) − Af−1,−2 (x) = ≥ 0.
18
The equality holds for x = y = z, and for −x = y = z (or any cyclic permuta-
tion).
The condition (a), namely f6 (x, 1, 1) − Af−1,−2 (x) ≥ 0 for x ∈ (−5, −1), is satis-
fied.
The condition (b) is satisfied if f6 (x, 1, 1) − Af−1,−1 (x) ≥ 0 for x ∈ (−∞, −5] ∪
[−1, ∞). We have
(x − 1)4 (x + 1)4
f−1,−1 (x) = ,
36(x + 2)2
(x 2 − 1)2 g(x)
f6 (x, 1, 1) − Af−1,−1 (x) = ,
8(x + 2)2
where
We need to show that g1 (x) ≤ 0 for x ≤ −5, and g1 (x) ≥ 0 for x ≥ −1. Indeed, if
x ≤ −5, then
g1 (x) < 5(3x 3 + 11x 2 + 37x + 54) = 5(x + 2)(3x 2 + 5x + 27) < 0,
The equality holds for x = y = z, and for −x = y = z (or any cyclic permuta-
tion).
15
(2x 2 + y 2 +z 2 )(2 y 2 +z 2 + x 2 )(2z 2 + x 2 + y 2 ) ≥ 64x 2 y 2 z 2 + (x − y)2 ( y −z)2 (z− x)2 .
4
(Vasile C., 2011)
Highest Coefficient Cancellation Method for Real Variables 267
The condition (a), namely f6 (x, 1, 1) − Af−1,−2 (x) ≥ 0 for x ∈ (−5, −1), is satis-
fied.
The condition (b) is satisfied if f6 (x, 1, 1) − Af−1,−1 (x) ≥ 0 for x ∈ (−∞, −5] ∪
[−1, ∞). We have
(x − 1)4 (x + 1)4
f−1,−1 (x) = ,
36(x + 2)2
(x 2 − 1)2 (x + 5)g1 (x)
f6 (x, 1, 1) − Af−1,−1 (x) = ≥ 0,
16(x + 2)2
where
g1 (x) = 15x 3 + 53x 2 + 185x + 227.
As shown at the preceding P 3.35, we have (x + 5)g1 (x) ≥ 0.
The equality holds for x = y = z, and for −x = y = z (or any cyclic permuta-
tion).
Observation 1. The inequalities in P 3.34, P 3.35 and P 3.36 are particular cases
of the following statement:
268 Vasile Cîrtoaje
We have
9(2k2 + 29k + 2)
A= ,
16
f6 (x, 1, 1) = (x 2 − 1)2 [k x 2 + 2(k + 1)2 ],
(x 2 − 1)2 (x + 5)g(x)
f6 (x, 1, 1) − Af−1,−2 (x) = ,
36
g(x) = (−2k2 + 7k − 2)x + 14k2 + 23k + 14.
turns into
A(k, δk )t 6 + B(k, δk )t 4 + 4C(k, δk )t 2 ≥ 0,
where
C(k, δk ) = 3(k + 2)(2k + 1) − 16δk .
From the necessary condition C(k, δk ) ≥ 0, we get
3(k + 2)(2k + 1)
δk ≤ .
16
Since ( (
f6 (x, y, z) = 8 (p2 − 2q + yz − x 2 ) − 3 (p2 − 2q − x 2 ),
Highest Coefficient Cancellation Method for Real Variables 269
So, we only need to show that there exists a real number δ such that
where
g(x) = 2(3x 2 + 2x + 3) − [x 2 + 5x + 8 + δ(x + 2)(x + 5]2 .
Choosing δ = −1, we get
Since
Indeed,
The equality holds for x = y = z, and for x/4 = y = z (or any cyclic permutation).
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 4, y = 1 + t and z = 1 − t, the inequality
(16x 2 + 3 yz)(x − y)(x − z)(x − 4 y)(x − 4z) + α(x − y)2 ( y − z)2 (z − x)2 ≥ 0
turns into
At 6 + B t 4 + C t 2 ≥ 0,
where C = 324(α − 52). The necessary condition C ≥ 0 involves α ≥ 52.
A = 3(1 + 2) + 27 = 36.
Because
f6 (x, 1, 1) = x 4 (x − 1)2 ,
x 4 (x − 1)4
f0,0 (x) = ,
36(x + 2)2
3x 4 (x − 1)2 (2x + 1)
f6 (x, 1, 1) − Af0,0 (x) = ,
(x + 2)2
272 Vasile Cîrtoaje
The condition (a), namely f6 (x, 1, 1) ≥ Af0,0 (x) for x > −1/2, is satisfied.
The condition (b) is satisfied if there is a real number γ such that
13 − 3 17 13 + 3 17
P 3.40. Let x, y, z be real numbers. If ≤k≤ , then
2 2
2
k
(x 2 + k yz)2 (x − y)(x − z) + − 1 (x − y)2 ( y − z)2 (z − x)2 ≥ 0.
4
(Vasile C., 2012)
Highest Coefficient Cancellation Method for Real Variables 273
Fγ,δ = fγ,−2 .
Since
4(x − 1)4 (x − γ)2
fγ,−2 (x) = ,
81(2 + γ)2
we have
(x − 1)2 f1 (x) f2 (x)
f6 (x, 1, 1) − Afγ,−2 (x) = ,
9(2 + γ)2
where
4(x − 1)4
f∞,−2 (x) = ,
81
Solution. Denote
m= −k, m ≥ 0.
Write the inequality as f6 (x, y, z) ≥ 0, where
k2
f6 (x, y, z) = (x 2 + k yz)2 (x − y)(x − z) + − 1 (x − y)2 ( y − z)2 (z − x)2 .
4
As shown at the preceding P 3.40, f6 (x, y, z) has the highest coefficient
9
A = (k + 4)2 .
4
Also,
f6 (x, 1, 1) = (x 2 + k)2 (x − 1)2 .
Case 1: k ≤ −4. The case k = −4 is treated in P 3.2. Further, consider k < −4.
Since
4(x − 1)4 (x − m)2 (x + m)2 (x 2 + k)2 (x − 1)2
f−m,m (x) = = ,
9(4 + 2m ) (x + 2)
2 2 2 9(2 − k)2 (x + 2)2
3(x 2 + k)2 (x − 1)2 [(8 − k)x + 4 − 5k][(−k)x + 4 − k]
f6 (x, 1, 1) − Af−m,m (x) = ,
4(2 − k)2 (x + 2)2
we apply Theorem 2 and Remark 3 for
5k − 4 (4 − k)2
and η = < −2, which involves ξ = and
8−k k(8 − k)
5k − 4 4 − k 5k − 4 4−k
= , , \ = −∞, ∪ ,∞ .
8−k k 8−k k
k − 20
β= ,
3(k + 4)
we have (see the preceding P 3.40)
where
g2 (x) = (17 − k)x 2 − 4(1 + k)x + 2(7k − 2).
Since k +1 < 0, we
must showthat g2 (x) ≤ 0 for x ∈ . Since 17 − k > 0, this is
5k − 4 4−k
true if g2 ≤ 0 and g2 ≤ 0. We have
8−k k
5k − 4 9(k + 1)(k + 4)2
g2 = < 0,
8−k (8 − k)2
4−k (k + 4)(17k2 − 59k + 68)
g2 = < 0.
k k2
Case 2: k ∈ [−1, 0]. Since the case k = −1 is treated in P 3.14, and the case
k = 0 in P 3.39, consider further k ∈ (−1, 0). The proof is similar to the one of the
case 1. We set
5k − 4 −1
η= ∈ −1, ,
8−k 2
we involves
4 − k 5k − 4 4−k 5k − 4
= , , \ = −∞, ∪ ,∞ .
k 8−k k 8−k
The condition (b) is clearly satisfied. Since k + 1 > 0, the condition (a) is satisfied
if g2 (x) ≥ 0 for
4 − k 5k − 4
x ∈= , .
k 8−k
Since g2 is decreasing, we have
5k − 4 9(k + 1)(k + 4)2
g2 (x) ≥ g3 = > 0.
8−k (8 − k)2
for x = y = z, for x/ −k = y = z (or any cyclic permutation)
The equality holds
if k = 0, for −x/ −k = y = z (or any cyclic permutation) if k = 0, and for x = 0
and y + z = 0 (or any cyclic permutation). If k = 0, then the equality holds also
for x = 0 and y = z (or any cyclic permutation).
Solution. Write the inequality as f6 (x, y, z) ≥ 0, where f6 (x, y, z) has the highest
coefficient (see P 3.40)
9
A = (k + 4)2 .
4
We have
f6 (x, 1, 1) = (x 2 + k)2 (x − 1)2 ,
and
(x 2 + k)2 (x − 1)2
f
−k,−
−k (x) = ,
9(2 − k)2 (x + 2)2
3(x 2 + k)2 (x − 1)2 [(8 − k)x + 4 − 5k][(−k)x + 4 − k]
f6 (x, 1, 1)−Af
−k,−
−k (x) = .
4(2 − k)2 (x + 2)2
Case 1: 0 ≤ k ≤ 2. The cases k = 2 and k = 0 are treated in P 3.22 and P 3.39,
respectively. Consider further 0 < k < 2. We choose
5k − 4 −1
η= ∈ ,1 ,
8−k 2
which involves
5k − 4 4 − k 5k − 4 4−k
= , , \ = −∞, ∪ ,∞ .
8−k k 8−k k
k − 20
δ= ,
3(k + 4)
If
13 − 3 17 13 + 3 17
≤k≤ ,
2 2
278 Vasile Cîrtoaje
then the inequality g2 (x) ≥ 0 holds for all x ∈ . So, we only need to show that
g2 (x) ≥ 0 for
13 − 3 17
0<k< ≈ 0.315.
2
5k − 4 4−k
Since g2 is decreasing on −∞, and increasing on , ∞ , we only
8−k k
5k − 4 4−k
need to show that g2 ≥ 0 and g2 ≥ 0. Indeed, we have
8−k k
5k − 4 9(k + 1)(k + 4)2
g2 = >0
8−k (8 − k)2
and
4−k (k + 4)(17k2 − 59k + 68)
g2 = > 0.
k k2
Thus, the proof is completed.
Case 2: 2 < k ≤ 8. The inequality is treated in P 3.40.
Case 3: k > 8. We choose
5k − 4
η= ∈ (−∞, −5),
8−k
which involves
5k − 4 4 − k 5k − 4 4−k
= , , \ = −∞, ∪ ,∞ .
8−k k 8−k k
k − 20
δ= ,
3(k + 4)
The inequalities in P 3.2, P 3.14, P 3.22 and P 3.39 are particular cases of this
general statement.
The coefficient of the product (x − y)2 ( y −z)2 (z − x)2 is the best possible. Setting
x = 1, y = 1 + t and z = 1 − t, the inequality
(x 2 + k yz)2 (x − y)(x − z) + αk (x − y)2 ( y − z)2 (z − x)2 ≥ 0.
turns into
At 6 + B t 4 + C t 2 ≥ 0,
k2
where A = 4αk + 4 − k2 . The necessary condition A ≥ 0 involves αk ≥ − 1.
4
f6 (0, 1, −1) = 2 · 2 − 4 = 0,
according to (3.19), we will choose
1 h (−2)
γ= + ,
3 12A
Highest Coefficient Cancellation Method for Real Variables 281
where
h(x) = f6 (x, 1, 1) = x 2 (x 2 + 8)(x − 1)2 .
We get
h (−2) −2 −1
h (−2) = −864, = , γ= ,
12A 3 3
therefore 2
1
h−1/3,δ = x − (x + 2)(2x + 1) + δ(x + 2)3 ,
3
f6 (x, 1, 1) − Ah−1/3,δ (x) = x 2 (x 2 + 8)(x − 1)2 − 12[−2x 2 − 2x − 2 + 3δ(x + 2)3 ]2 .
The inequality f6 (x, 1, 1) − Ah−1/3,δ (x) ≥ 0 holds in the vicinity of −2 only for
δ = 5/54. Setting this value for δ, we get
2
5
f6 (x, 1, 1) − Ah−1/3,δ (x) = x 2 (x 2 + 8)(x − 1)2 − 12 −2x 2 − 2x − 2 + (x + 2)3
18
1
= x 2 (x 2 + 8)(x − 1)2 − (5x 3 − 6x 2 + 24x + 4)2
27
1
= (2x 6 + 6x 5 − 33x 4 − 184x 3 − 312x 2 − 192x − 16)
27
1
= (x + 2)4 (2x 2 − 10x − 1) ≥ 0.
27
The equality holds for x = y = z, for x = 0 and y = z (or any cyclic permutation),
and for x = 0 and y + z = 0 (or any cyclic permutation).
Observation 1. From the inequalities in P 3.39 and P 3.43, we get the following
inequality:
x 2 (x 2 + k yz)(x − y)(x − z) ≥ (x − y)2 ( y − z)2 (z − x)2 , 0 ≤ k ≤ 8.
Indeed, since the left hand side is linear in k, it suffices to prove this inequality only
for k = 0 (see P 3.39) and k = 8 (see P 3.43).
Observation 2. Notice that the coefficient of the product (x − y)2 ( y − z)2 (z − x)2
in the inequality from Observation 1 is the best possible. Indeed, setting x = 0,
y = 1 + t and z = 1 − t, the inequality
x 2 (x 2 + k yz)(x − y)(x − z) ≥ α(x − y)2 ( y − z)2 (z − x)2
turns into
At 4 + B t 2 + C ≥ 0,
where A = 1 − α. The necessary condition A ≥ 0 involves α ≤ 1.
282 Vasile Cîrtoaje
Solution. Let
f (x, y, z) = (x − y)(x − z)(x − 2 y)(x − 2z)(2x − y)(2x − z)
and
f6 (x, y, z) = f (x, y, z) + 15(x − y)2 ( y − z)2 (z − x)2 .
Using the identities
(x − y)(x − z) = x 2 + 2 yz − q,
(x − 2 y)(x − 2z) = x 2 + 6 yz − 2q,
(2x − y)(2x − z) = 4x 2 + 3 yz − 2q,
Since
f6 (x, 1, 1) = (x − 1)2 (x − 2)2 (2x − 1)2 ,
4(x − 1)4 (x − 2)2 (2x − 1)2
f2,1/2 (x) = ,
9(x + 2)2
3(x − 1)2 (x − 2)2 (2x − 1)2 (5x − 2)(2 − x)
f6 (x, 1, 1) − Af2,1/2 (x) = ,
(x + 2)2
apply Theorem 2 for
Eα,γ = f2,1/2 , Fγ,δ = f2,−2 ,
and for η = 2, which implies ξ = 16/5 and
2 2
= ,2 , \ = −∞, ∪ [2, ∞).
5 5
2
The condition (a), namely f6 (x, 1, 1) ≥ Af2,1/2 (x) for x ∈ , 2 , is satisfied.
5
Highest Coefficient Cancellation Method for Real Variables 283
2
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Af2,−2 (x) for x ∈ −∞, ∪[2, ∞).
5
From
(x − 1)4 (x − 2)2
f2,−2 (x) = ,
324
it follows that this inequality is equivalent to
which is true.
The equality holds for x = y = z, for x/2 = y = z (or any cyclic permutation),
and for 2x = y = z (or any cyclic permutation).
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = y + z in
(x − y)(x − z)(x − 2 y)(x − 2z)(2x − y)(2x − z) + α(x − y)2 ( y − z)2 (z − x)2 ≥ 0,
we get
( y − z)2 [(α − 15) y 2 z 2 + 4( y 2 + z 2 )( y − z)2 ] ≥ 0,
which holds for all real numbers y and z only if α ≥ 15.
Solution. Let
f (x, y, z) = (x − y)(x − z)(x − 2 y)(x − 2z)(x − 3 y)(x − 3z),
and
f6 (x, y, z) = f (x, y, z) − 3(x − y)2 ( y − z)2 (z − x)2 .
From
(x − y)(x − z) = x 2 + 2 yz − q,
(x − 2 y)(x − 2z) = x 2 + 6 yz − 2q,
(x − 3 y)(x − 3z) = x 2 + 12 yz − 3q,
Since
f6 (x, 1, 1) = (x − 1)2 (x − 2)2 (x − 3)2 ,
4(x − 1)4 (x − 2)2 (x − 3)2
f2,3 (x) = ,
1521(x + 2)2
3(x − 1)2 (x − 2)2 (x − 3)2 (2 + 11x)(46 − 7x)
f6 (x, 1, 1) − Af2,3 (x) = ,
169(x + 2)2
apply Theorem 2 for
Eα,γ = f2,3 , Fγ,δ = fγ,−2 ,
and for η = 46/7, which implies ξ = 400/77 and
−2 46 −2 46
= , , \ = −∞, ∪ ,∞ .
11 7 11 7
−2 46
The condition (a), namely f6 (x, 1, 1) ≥ Af2,3 (x) for x ∈ , , is satisfied.
11 7
The condition (b) is satisfied if there exists a real γ such that f6 (x, 1, 1) ≥ Afγ,−2 (x)
−2 46
for x ∈ −∞, ∪ , ∞ . We have
11 7
4(x − 14 (x − γ)2
fγ,−2 (x) = ,
81(2 + γ)2
(x − 1)2 g(x)
f6 (x, 1, 1) − Afγ,−2 (x) = ,
9(2 + γ)2
where
g(x) = 9(2 + γ)2 (x − 2)2 (x − 3)2 − 400(x − 1)2 (x − γ)2 = g1 (x)g2 (x),
to get
40
g1 (x) = (x + 2)2 ≥ 0,
7
40
g2 (x) = (8x 2 − 49x + 50) > 0.
7
The equality holds for x = y = z, for x/2 = y = z (or any cyclic permutation),
for x/3 = y = z (or any cyclic permutation), and for x = 0 and y + z = 0 (or any
cyclic permutation).
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, y = 1 and z = −1 in the inequality
(x − y)(x − z)(x − 2 y)(x − 2z)(x − 3 y)(x − 3z) ≥ α(x − y)2 ( y − z)2 (z − x)2
involves α ≤ 3.
and
f6 (x, y, z) = f (x, y, z) + 15(x − y)2 ( y − z)2 (z − x)2 .
From
(x − y)(x − z) = x 2 + 2 yz − q,
(2x + 3 y)(2x + 3z) = 4x 2 + 3 yz + 6q,
(3x + 2 y)(3x + 2z) = 9x 2 − 2 yz + 6q,
Since
f6 (x, 1, 1) = (x − 1)2 (2x + 3)2 (3x + 2)2 ,
4(x − 1)4 (2x + 3)2 (3x + 2)2
f−3/2,−2/3 (x) = ,
9 · 625(x + 2)2
3(x − 1)2 (2x + 3)2 (3x + 2)2 (7x + 18)(29x + 46)
f6 (x, 1, 1) − Af−3/2,−2/3 (x) = ,
625(x + 2)2
apply Theorem 2 for η = −46/29, which implies ξ = −16/203 and
−18 −46 −18 −46
= , , \ = −∞, ∪ ,∞ ,
7 29 7 29
and for
Eα,γ = fα,−2 , Fγ,δ = f−3/2,−2/3 .
The condition
(a)is satisfied if there exists a real α such that f6 (x, 1, 1) ≥ Afα,−2 (x)
−18 −46
for x ∈ , . We have
7 29
4(x − 14 (x − α)2
fα,−2 (x) = ,
81(2 + α)2
(x − 1)2 g(x)
f6 (x, 1, 1) − Afα,−2 (x) = ,
9(2 + α)2
where
g(x) = 9(2 + α)2 (2x + 3)2 (3x + 2)2 − 16(x − 1)2 (x − α)2 .
It is easy to check that g(−2) = 0. Choosing α = −46/29, the inequality g(x) ≥ 0
is equivalent to
(x + 2)2 (83x 2 + 134x + 8) ≥ 0,
which is true because 83x 2 + 134x + 8 > 0 for x ≤ −46/29.
−18
The condition (b), namely f6 (x, 1, 1) ≥ Af−3/2,−2/3 (x) for x ∈ −∞, ∪
7
−46
, ∞ , is satisfied.
29
The equality holds for x = y = z, for −3x/2 = y = z (or any cyclic permutation),
for −2x/3 = y = z (or any cyclic permutation), and for x = 0 and y + z = 0 (or
any cyclic permutation).
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, y = 1 and z = −1 in
(x − y)(x −z)(2x +3 y)(2x +3z)(3x +2 y)(3x +2z)+α(x − y)2 ( y −z)2 (z− x)2 ≥ 0,
yields α ≥ 15.
Highest Coefficient Cancellation Method for Real Variables 287
Solution. Let
f (x, y, z) = (x + y)2 (x + z)2 (x − y)(x − z)
and
1
f6 (x, y, z) = f (x, y, z) + (x − y)2 ( y − z)2 (z − x)2 .
4
Since
(x + y)(x + z) = x 2 + q, (x − y)(x − z) = x 2 + 2 yz − q,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = x 4 (x 2 + 2 yz),
1 9
A = 9 + (−27) = .
4 4
On the other hand, since
(x + 1)4 (x − 1)4
f−1,−1 (x) = ,
36(x + 2)2
(x + 1)4 (x − 1)4
Af−1,−1 (x) = ,
16(x + 2)2
3(x + 1)4 (x − 1)2 (x + 3)(5x + 7)
f6 (x, 1, 1) − Af−1,−1 (x) = ,
16(x + 2)2
we will apply Theorem 2 for η = −3, which involves ξ = −1/55,
−7 −7
= −3, , \ = (−∞, −3] ∪ ,∞ ,
5 5
and for
Eα,β = fα,−2 , Fγ,δ = f−1,−1 .
288 Vasile Cîrtoaje
−7
Condition (a): x ∈ = −3, . We need to show that there exists a real α
5
such that
f6 (x, 1, 1) ≥ Afα,−2 (x).
Since
4(x − 1)4 (x − α)2
fα,−2 (x) = ,
81(α + 2)2
(x − 1)4 (x − α)2
Afα,−2 (x) = ,
9(α + 2)2
we have
g1 (x)g2 (x)
f6 (x, 1, 1) − Afα,−2 (x) = ,
9(α + 2)2
where
then
(x − y)(x − z)(x − k y)2 (x − kz)2 + αk (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
−2 ≤ k < 0.
We have
9
A = (k + 2)4 ,
4
f6 (x, y, z) = (x − 1)2 (x − k)2 ,
(x − 1)4 (x − k)4
f k,k (x) = ,
9(k − 1)2 (k + 2)2
(x − 1)2 (x − k)4
f6 (x, 1, 1) − Af k,k (x) = h(x),
4(k − 1)2 (x + 2)2
2−k (k + 2)2
η= ≤ −2, ξ= ,
k k(4 − k)
2 − k 5k − 2 2−k 5k − 2
= , , \ = −∞, ∪ ,∞ ,
k 4−k k 4−k
and for
Eα,β = fα,−2 , Fγ,δ = f k,k .
2−k 5k − 2
Condition (b): x ∈ \ = −∞, ∪ , ∞ . Since h(x) ≥ 0, we
k 4−k
have f6 (x, 1, 1) − Af k,k (x) ≥ 0.
2 − k 5k − 2
Condition (a): x ∈ = , . By choosing
k 4−k
5k − 2
α= ,
4−k
we get
(x − 1)2
f6 (x, 1, 1) − Afα,−2 (x) = g1 (x)g2 (x),
9
290 Vasile Cîrtoaje
Solution. Let
f (x, y, z) = (x − y)(x − z)(x − 4 y)2 (x − 4z)2
and
f6 (x, y, z) = f (x, y, z) − 39(x − y)2 ( y − z)2 (z − x)2 .
Since
A1 = P1 (1, 1, 1) = 3969
(x − 1)4 (x − 4)4
f4,4 (x) = ,
2916(x + 2)2
(x − 1)4 (x − 4)4
Af4,4 (x) = ,
(x + 2)2
3(x − 1)2 (x − 4)4 (2x + 1)
f6 (x, 1, 1) − Af4,4 (x) = ,
(x + 2)2
we will apply Theorem 2 for η → ∞, which involves ξ → ∞,
−1 −1
= ,∞ , \ = −∞, ,
2 2
and for
Eα,β = f4,4 Fγ,δ = f∞,−2 .
−1
Condition (a): x ∈ = , ∞ . Because
2
f6 (x, 1, 1) − Af4,4 (x) ≥ 0,
4(x − 1)4
f∞,−2 (x) = ,
81
Af∞,−2 (x) = 144(x − 1)4 ,
f5 (x, 1, 1) − Af∞,−2 (x) = (x − 1)2 (x + 2)2 (x 2 − 20x + 29) ≥ 0.
The equality holds for x = y = z, for x/4 = y = z (or any cyclic permutation),
and for x = 0 and y + z = 0 (or any cyclic permutation).
Observation 1. Similarly, we can prove the following generalization:
• Let x, y, z be real numbers. If
1
αk = (k2 − 2k − 2)(k2 + 2k + 2), 1 ≤ k ≤ 4,
4
then
(x − y)(x − z)(x − k y)2 (x − kz)2 + αk (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
2 − k 2(1 + 7k + k2 ) 5k − 2
< <
k 17 + 2k − k2 4−k
and
2−k 5k − 2
g3 > 0, g3 > 0.
k 4−k
k ∈ (0, 1).
2−k (k + 2)2
η= ∈ (1, ∞), ξ= ,
k k(4 − k)
5k − 2 2 − k 5k − 2 2−k
= , , \ = −∞, ∪ ,∞ .
4−k k 4−k k
We have g3 (x) > 0 for x ∈ \ and 0 < k < 1, because
5k − 2 2(1 + 7k + k2 ) 2 − k
< <
4−k 17 + 2k − k2 k
and
5k − 2 2−k
g3 > 0, g3 > 0.
4−k k
294 Vasile Cîrtoaje
Observation 3. From P 3.47, P 3.48 and the observations attached to them, the
following generalization follows:
1
αk = (k2 − 2k − 2)(k2 + 2k + 2), k ∈ ,
4
then
(x − y)(x − z)(x − k y)2 (x − kz)2 + αk (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
Then,
x 2 (x − y)(x − z)(x − k y)(x − kz) ≥ αk (x − y)2 ( y − z)2 (z − x)2 .
Solution. Denote
f (x, y, z) = x 2 (x − y)(x − z)(x − k y)(x − kz),
Since
Therefore, since the product (x − y)2 ( y − z)2 (z − x)2 has the highest coefficient
equal to −27, f6 (x, y, z) has the highest coefficient
⎧
⎪ 9(2 + k)2 (4 − k)2
⎪
⎪ , k ∈ (−∞, −2] ∪ [4, ∞)
⎨ 4(1 − k)2
A = A1 + 27αk = 9(2 + k)2 , k ∈ [−2, 1] .
⎪
⎪
⎪
⎩
9(4 − k)2 , k ∈ [1, 4]
In addition, we have
k ∈ {−2, 4}, k = 0,
k ∈ (−∞, −2), k ∈ (4, ∞), k ∈ (−2, 0), k ∈ (0, 1], k ∈ [1, 4).
Case 1: k ∈ {−2, 4}. Since A = 0 and f6 (x, 1, 1) ≥ 0 for any real x, the conclusion
follows by Corollary 1.
4 4(k + 2)2
η= ∈ [4, ∞), ξ= ,
k k(k + 8)
4k − 4 4 4k − 4 4
= , , \ = −∞, ∪ ,∞ ,
k+8 k k+8 k
and for
Eα,β = f k,0 , Fγ,δ = fγ,−2 .
4k − 4 4
The condition (a), namely f6 (x, 1, 1) ≥ Af k,0 (x) for x ∈ , , is satisfied.
k+8 k
4k − 4
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Afγ,−2 (x) for x ∈ −∞, ∪
k+8
4
, ∞ . This is true if g1 (x)g2 (x) ≥ 0, where
k
3x 3 (x − 1)2 (x − k)2 (4 − x)
f6 (x, 1, 1) − Af k,0 (x) = ,
(x + 2)2
we apply Theorem 2 for
η = 4, ξ = 4,
= (0, 4), \ = (−∞, 0] ∪ [4, ∞),
and for
Eα,β = f k,0 , Fγ,δ = f0,−2 .
The condition (a), namely f6 (x, 1, 1) ≥ Af k,0 (x) for x ∈ (0, 4), is satisfied.
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈ (−∞, 0] ∪ [4, ∞).
We have
for x ≥ 4.
The proof is completed. The equality hold for x = y = z, for x = 0 and y = z
(or any cyclic permutation), for x/k = y = z (or any cyclic permutation) if k = 0,
and for x = 0 and y + z = 0 (or any cyclic permutation) if k ∈ [−2, 1].
Observation 1. The coefficient αk of the product (x − y)2 ( y −z)2 (z − x)2 is the best
possible. Setting x = k, y = 1 + t and z = 1 − t, the inequality in P 3.49 becomes
where
A(k, αk ) = 4(1 + k − αk ),
C(k, αk ) = (1 − k)2 [20 + 12k − 4k2 − k4 − 4(1 − k)2 αk ].
20 + 12k − 4k2 − k4
αk ≤ 1 + k, αk ≤ ,
4(1 − k)2
300 Vasile Cîrtoaje
( y − z)2 [ y 4 + z 4 − (k − 1) yz( y 2 + z 2 ) + (1 − k − αk ) y 2 z 2 ] ≥ 0.
y 4 + z 4 − (k − 1) yz( y 2 + z 2 ) + (1 − k − αk ) y 2 z 2 ≥ 0,
we get
αk ≤ 5 − 3k.
Observation 2. There are some relevant particular cases of the inequality in P 3.49.
• For k = −2, the inequality turns into
x 2 (x − y)(x − z)(x + 2 y)(x + 2z) + (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
which is equivalent to
" #
(x + y + z)2 x 4 + y 4 + z 4 + x yz(x + y + z) − x y(x 2 + y 2 ) ≥ 0.
which is equivalent to
" #2
x 3 + y 3 + z 3 + 3x yz − x y(x + y) ≥ 0.
!
The equality occurs for x 3 + y 3 + z 3 + 3x yz = x y(x + y).
• For k = 4, the inequality turns into
x 2 (x − y)(x − z)(x − 4 y)(x − 4z) + 7(x − y)2 ( y − z)2 (z − x)2 ≥ 0,
which is equivalent to
(x 2 + y 2 + z 2 − x y − yz − z x)(x 2 + y 2 + z 2 − 2x y − 2 yz − 2z x)2 ≥ 0.
The equality occurs for x = y = z, and for x= y+ z (or any cyclic permuta-
tion).
Observation 3. Substituting k − 1 for k and using then the identity
x 2 (x − y)(x − z)[x − (k − 1) y][x − (k − 1)z] =
Highest Coefficient Cancellation Method for Real Variables 301
= x 2 (x − y)(x − z)(x − k y + z)(x − kz + y) + k(x − y)2 ( y − z)2 (z − x)2 ,
we get the following statement:
• Let x, y, z be real numbers, and let
⎧
⎪ (3 − k)(1 + k)3
⎪
⎪ , k ∈ (−∞, −1] ∪ [5, ∞)
⎨ 4(2 − k)2
α∗k = 0, k ∈ [−1, 2] .
⎪
⎪
⎪
⎩
4(2 − k), k ∈ [2, 5]
Then,
x 2 (x − y)(x − z)(x − k y + z)(x − kz + y) ≥ α∗k (x − y)2 ( y − z)2 (z − x)2 ,
with equality for x = y = z, for x = 0 and y = z (or any cyclic permutation), for
x/(k − 1) = y = z (or any cyclic permutation) if k = 1, and for x = 0 and y + z = 0
(or any cyclic permutation) if k ∈ [−1, 2].
the inequality follows from Corollary 1. Notice that this case is treated in P 3.17.
−1
Case 3: k ∈ −2, . Since
2
9(k + 2)2 (2k + 1)2 (k + 2)2 (x − 1)4 (x − k)2
A= , Af k,∞ (x) = ,
4(k − 1)2 (k − 1)2 (x + 2)2
−1
The condition (a), namely f6 (x, 1, 1) − Af k,∞ (x) ≥ 0 for x > , is satisfied.
2
−1
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Af∞,−2 (x) for x ≤ . Indeed, we
2
have
(2k + 1)2 (x − 1)4
Af∞,−2 (x) = ,
9
304 Vasile Cîrtoaje
where
A(k, βk ) = 4βk − k2 ,
C(k, βk ) = (1 − k)2 [4(1 − k)2 βk + k(8 + 11k + 8k2 )].
From the necessary conditions A(k, βk ) ≥ 0 and C(k, βk ) ≥ 0, we get
yz + βk ( y − z)2 ≥ 0.
1
Setting y = 1 and z = −1, we get βk ≥ .
4
Observation 2. There are some relevant particular cases of the inequality in P 3.50.
• For k = −2, the inequality turns into
yz(x − y)(x − z)(x + 2 y)(x + 2z) + (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
which is equivalent to
(x + y + z)2 x 2 ( y − z)2 + y 2 (z − x)2 + z 2 (x − y)2 ≥ 0.
The equality holds for x = y = z, for x + y + z = 0, and for y = z = 0 (or any cyclic
permutation).
Highest Coefficient Cancellation Method for Real Variables 305
which is equivalent to
(x y + yz + z x)2 (x 2 + y 2 + z 2 − x y − yz − z x) ≥ 0.
with equality for x = y = z, for y = z = 0 (or any cyclic permutation), for x/k =
y = z (or any cyclic permutation) if k = 0, and for x = 0 and y + z = 0 (or any cyclic
permutation) if k ∈ (−∞, −2] ∪ [1, ∞). If k = 0, then the equality holds also for
x = 0 and y = z (or any cyclic permutation).
with equality for x = y = z, for y = z = 0 (or any cyclic permutation), for x/k =
y = z (or any cyclic permutation) if k = 0, and for x = 0 and y + z = 0 (or any cyclic
permutation) if k ∈ (−∞, −2] ∪ [1, ∞). If k = 0, then the equality holds also for
x = 0 and y = z (or any cyclic permutation).
with equality for x = y = z, for y = z = 0 (or any cyclic permutation), for x/(k−1) =
y = z (or any cyclic permutation) if k = 1, and for x = 0 and y + z = 0 (or any cyclic
permutation) if k ∈ (−∞, −1] ∪ [2, ∞). If k = 1, then the equality holds also for
x = 0 and y = z (or any cyclic permutation).
Highest Coefficient Cancellation Method for Real Variables 307
Solution. Denote
f (x, y, z) = (x 2 − y 2 )(x 2 − z 2 )(x − k y)(x − kz),
Since
(x + y)(x + z) = x 2 + q, (x − y)(x − z) = x 2 + 2 yz − q,
(x − k y)(x − kz) = x 2 + (k + k2 ) yz − kq,
f (x, y, z) has the same highest coefficient A1 as
P1 (x, y, z) = x 2 (x 2 + 2 yz)[x 2 + (k + k2 ) yz],
k ∈ {−5, −2}, k ∈ (−∞, −5) ∪ [1, ∞), k ∈ (−5, −2), k ∈ (−2, 1).
Case 1: k ∈ {−5, −2}. Since A = 0 and f6 (x, 1, 1) ≥ 0 for any real x, the conclusion
follows by Corollary 1.
The condition (b), namely f6 (x, 1, 1)−Af−1,k (x) ≥ 0 for x ∈ (−∞, −5]∪[−1, ∞),
is satisfied.
The condition (a) is satisfied if f6 (x, 1, 1) ≥ Af−1,−2 (x) for x ∈ (−5, −1). We
have
(k + 5)2 (x − 1)4 (x + 1)2
Af−1,−2 (x) = ,
36
(k − 1)(x + 5)(x 2 − 1)2 [7k + 5 − (k + 11)x]
f6 (x, 1, 1) − Af−1,−2 (x) = .
36
Thus, it suffices to show that 7k + 5 − (k + 11)x ≤ 0 for k ∈ (−∞, −5), and 7k +
5 − (k + 11)x ≥ 0 for k ∈ [1, ∞). Indeed, for k ∈ (−∞, −5), we have
4−k (k + 2)2
η= , ξ= ,
2k + 1 2k + 1
4−k 4−k
= k, , \ = (−∞, k] ∪ ,∞ ,
2k + 1 2k + 1
and for
Eα,β = f k,−2 , Fγ,δ = f−1,k .
4−k
The condition (b), namely f6 (x, 1, 1)−Af−1,k (x) ≥ 0 for x ∈ (−∞, k]∪ ,∞ ,
2k + 1
is satisfied.
4−k
The condition (a) is satisfied if f6 (x, 1, 1) ≥ Af k,−2 (x) for x ∈ k, . We
2k + 1
have
(k + 5)2 (x − 1)4 (x − k)2
Af k,−2 (x) =
9(k − 1)2
and
because
4−k 3(k − 1)(k + 5)
2k + 1 + (k − 4)x > 2k + 1 + (k − 4) · = > 0,
2k + 1 2k + 1
4−k
k − 4 + (2k + 1)x > k − 4 + (2k + 1) · = 0.
2k + 1
Case 4: k ∈ (−2, 1). Since
where
−(k + 2)2
η = −k − 4, ξ= ,
2k + 7
−k − 8 −k − 8
= −k − 4, , \ = (−∞, −k − 4] ∪ ,∞ .
2k + 7 2k + 7
and for
Eα,β = fα,−2 , Fγ,δ = f−1,k .
−k − 8
α=
2k + 7
to get
where
g3 (x) = (k + 8)x 2 + 5(1 − k)x − 5k − 4.
Since
−k − 8
g3 (x) = 2(k + 8)x + 5(1 − k) < 2(k + 8) ·
+ 5(1 − k)
2k + 7
−3(4k + 19k + 31) −3(2k + 19k + 30)
2 2
= <
2k + 7 2k + 7
−3(k + 2)(2k + 15)
= < 0,
2k + 7
g3 is strictly decreasing on , therefore
−k − 8 9(1 − k)(k + 2)2
g3 (x) > g3 = > 0.
2k + 7 (2k + 7)2
The proof is completed. The equality holds for x = y = z, for x/k = y = z (or
any cyclic permutation) if k = 0, for −x = y = z (or any cyclic permutation), and
for x = 0 and y + z = 0 (or any cyclic permutation) if k ∈ [−2, 1]. If k = 0, then
the equality holds also for x = 0 and y = z (or any cyclic permutation).
k2 (k + 1)(5k − 3)
γk ≥ , γk ≥ .
4 16
312 Vasile Cîrtoaje
Observation 2. There are some relevant particular cases of the inequality in P 3.51.
• For k = 1, the inequality turns into
1
(x + y)(x + z)(x − y)2 (x − z)2 + (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
4
which is equivalent to
2 2
x (2x − y − z) + y 2 (2 y − z − x) + z 2 (2z − x − y) ≥ 0.
which is equivalent to
(x + y + z)2 (x 4 + y 4 + z 4 − x 2 y 2 − y 2 z 2 − z 2 x 2 ) ≥ 0.
which is equivalent to
(x 2 + y 2 + z 2 + 3x y + 3 yz + 3z x)2 (x 2 + y 2 + z 2 − x y − yz − z x) ≥ 0.
Then,
(x 2 − y 2 )(x 2 − z 2 )(x − k y + z)(x − kz + y) + γ∗k (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
The condition (a), namely f6 (x, 1, 1) − Af∞,−2 (x) ≥ 0 for x > −1/2, is satisfied.
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Ah0,δ (x) for x ≤ −1/2. We have
The equality holds for x = y = z, and for y = z = 0 (or any cyclic permutation).
Observation 1. Similarly, we can prove the following generalization:
• Let x, y, z be real numbers. If
1 13 + 3 33
k∈ , k1 , k1 = ≈ 0.9448,
2 32
then
1
yz(x 2 + k yz)(x − y)(x − z) + (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
4
with equality for x = y = z, and also for y = z = 0 (or any cyclic permutation).
For
1
f6 (x, y, z) = yz(x 2 + k yz)(x − y)(x − z) + (x − y)2 ( y − z)2 (z − x)2 ,
4
we have
9
A = (4k + 1),
4
f6 (x, 1, 1) = (x 2 + k)(x − 1)2 ,
where
g(x) = (2 + 13k − 16k2 )x 2 − 2(5 + k − 4k2 )x − 1 + 25k − 28k2
= (2 + 13k − 16k2 )(x 2 + 2) − (5 + k − 4k2 )(2x + 1) ≥ 0.
13 + 3 33
k ∈ [0, k1 ] , k1 = ≈ 0.9448,
32
then
1
yz(x 2 + k yz)(x − y)(x − z) + (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
4
with equality for x = y = z, and also for y = z = 0 (or any cyclic permutation).
Since the left hand side of the inequality is linear in k, the inequality holds for
k ∈ [0, k1 ] if and only if it holds for k = 0 and k = k1 . These cases are treated in P
3.50 and Observation 1, respectively.
Observation 3. Replacing x, y, z with 1/x, 1/ y, 1/z, respectively, the statement
from Observation 2 becomes as follows:
• Let x, y, z be real numbers. If
13 + 3 33
k ∈ [0, k1 ] , k1 = ≈ 0.9448,
32
then
1
yz(k x 2 + yz)(x − y)(x − z) + (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
4
with equality for x = y = z, and also for y = z = 0 (or any cyclic permutation).
where
g(x) = x 4 + 2x 3 − x 2 + 6.
Since
f6 (−1, 1, 1) = 0, f6 (0, 1, −1) = 0,
we apply Corollary 2 for Fγ,δ = g−1,δ , where δ is given by (3.16):
for x ∈ , where
We have
g (x) = 4x 3 + 6x 2 − 2x, g (−2) = −4, δ = −1,
ḡ−1,−1 (x) = [x + 5x + 8 − (x + 2)(x + 5)] = 4(x + 1)2 ,
2 2
The equality holds for −x = y = z (or any cyclic permutation), and for x = 0
and y + z = 0 (or any cyclic permutation).
Highest Coefficient Cancellation Method for Real Variables 317
that is
A = P3 (1, 1, 1) = −9(−2 − 3)3 = 1125.
Since
P(x, 1, 1) = 4x 2 − 6x + 5,
f6 (x, 1, 1)
= (4x 2 − 6x + 5)[(2x 2 − 3x + 2) + 2] − 9(2x 2 − 3x + 2)
2x 2 − 3x + 2
= 2(4x 4 − 12x 3 + 13x 2 − 6x + 1) = 2(x − 1)2 (2x − 1)2 ,
The equality holds for x = y = z, and for 2x = y = z (or any cyclic permutation).
Observation. Similarly, we can prove the following generalization:
318 Vasile Cîrtoaje
From
f6 (x, y, z) =(k + 1) (k x 2 + 2 yz)[(k − 1) y 2 + p2 − 2q][(k − 1)z 2 + p2 − 2q]
(
− (k − 1) [(k − 1)x 2 + p2 − 2q],
that is
We have
where
g(x) = 2(k x 2 + 2)(x 2 + k + 1).
Since
f6 (−k − 1, 1, 1) = 0, f6 (0, 1, −1) = 0,
we apply Corollary 2 for Fγ,δ = g−k−1,δ , where δ is given by (3.16):
γ (γ + 2)2 g (−2)
δ= +
3(γ + 2) 12A
k+1 g (−2)
= + .
3(k − 1) 12(k + 5)(2k + 1)
320 Vasile Cîrtoaje
for x ∈ , where
2
γx 2 + γ(γ + 6)x − 8 δ(x + 2)(2γx + x + γ − 4)
ḡ−k−1,δ (x) = +
(γ + 2)3 (γ + 2)2
(k + 1)x 2 − (k + 1)(k − 5)x + 8 δ(x + 2)(2k x + x + k + 5)
= − .
(k − 1)3 (k − 1)2
We have
−k2 + 10k + 3
δ= ,
(k − 1)(k + 5)(2k + 1)
4[(2k + 1)x 2 − (k2 + k − 2)x + k + 5]2
ḡ−k−1,δ (x) = ,
(k − 1)2 (k + 5)2 (2k + 1)2
4[(2k + 1)x 2 − (k2 + k − 2)x + k + 5]2
Aḡ−k−1,δ (x) = ,
(k + 5)(2k + 1)
2g1 (x)
g(x) − Aḡ−k−1,δ (x) = ,
(k + 5)(2k + 1)
The equality holds for −x/(k + 1) = y = z (or any cyclic permutation), and for
x = 0 and y + z = 0 (or any cyclic permutation).
1 1 1 1
+ + + ≤ 0.
3x 2 + y 2 + z 2 3 y 2 + z 2 + x 2 3z 2 + x 2 + y 2 x y + yz + z x
Since ( (
(3x 2 + y 2 + z 2 ) =
(2x 2 + p2 − 2q),
+
f6 (x, y, z) has the same highest coefficient as (2x 2 ), that is
A = 8.
We have
Since
f6 (0, 1, −1) = 0,
we apply Corollary 2 for Fγ,δ = hγ,δ , where γ is given by (3.19):
1 h (−2) 1 h (−2)
γ= + = + ,
3 12A 3 96
with
We have
h (x) = 2(9x 5 + 35x 4 + 66x 3 + 108x 2 + 100x + 32),
1 1 1
h (−2) = 16, γ= + = ,
3 6 2
1
h1/2,δ (x) = [x + (x + 2)(2x + 1) + δ(x + 2)3 ]2 .
2
Choosing
−1
δ= ,
4
we get
2
1 1
h1/2,−1/4 (x) = x + (x + 2)(2x + 1) − (x + 2)3
2 4
1 3
= (x + 2x 2 − 2x + 4)2 ,
16
1
f6 (x, 1, 1) − Ah1/2,−1/4 (x) = (x + 2)2 (5x 4 + 4x 3 + 30x 2 + 8x + 28).
2
322 Vasile Cîrtoaje
For x ≥ 0, the condition f6 (x, 1, 1) − Ah1/2,−1/4 (x) ≥ 0 is clearly true. It is also true
for x < 0, since
x 2 − yz y2 − z x z2 − x y
+ + 2 ≤ 1.
3x 2 + y +z
2 2 3y + z + x
2 2 2 3z + x 2 + y 2
(Vasile C., 2011)
Solution. Write the inequality as f6 (x, y, z) ≥ 0, where
(
f6 (x, y, z) = (3x 2 + y 2 + z 2 ) − (x 2 − yz)(3 y 2 + z 2 + x 2 )(3z 2 + x 2 + y 2 ).
From
(
f6 (x, y, z) = (2x 2 + p2 − 2q) − (x 2 − yz)(2 y 2 + p2 − 2q)(2z 2 + p2 − 2q),
A = −4 + 12 = 8.
We have
f6 (0, 1, −1) = 16 − 16 = 0,
Since
f6 (0, 1, −1) = 0,
we apply Corollary 2 for Fγ,δ = hγ,δ , where γ is given by (3.19):
1 h (−2) 1 h (−2)
γ= + = + ,
3 12A 3 96
Highest Coefficient Cancellation Method for Real Variables 323
with
h(x) = (x 2 + 4)(2x 4 + 6x 3 + 5x 2 + 4x + 8)
= 2x 6 + 6x 5 + 13x 4 + 28x 3 + 28x 2 + 16x + 32.
We have
h (x) = 2(6x 5 + 15x 4 + 26x 3 + 42x 2 + 28x + 8),
1 5 −1
h (−2) = −80, γ= − = ,
3 6 2
1
h−1/2,δ (x) = [x − (x + 2)(2x + 1) + δ(x + 2)3 ]2 .
2
Choosing
δ = 0,
we get
2
1
h−1/2,0 (x) = x − (x + 2)(2x + 1)
2
1
= (2x 2 + 3x + 2)2 ,
4
From
(
f6 (x, y, z) = 2 yz[(k−1) y 2 +p2 −2q][(k−1)z 2 +p2 −2q]+ [(k−1)x 2 +p2 −2q],
that is
A = 6(k − 1)2 + (k − 1)3 = (k − 1)2 (k + 5) ≥ 0.
We have
Since
f6 (0, 1, −1) = 0,
we apply Corollary 2 for Fγ,δ = hγ,0 , where γ is given by (3.19):
1 h (−2) 1 h (−2)
γ= + = + ,
3 12A 3 12(k − 1)2 (k + 5)
with
We have
g(x) = k(k + 5)x 4 + 2(k3 + 2k2 + 9k − 8)x 2 − 8(k2 − k + 2)x + k3 + 6k2 + 17k − 4.
Since
k3 + 2k2 + 9k − 8 ≥ 3k2 + 9k − 8 > 2(k2 − k + 2)
Highest Coefficient Cancellation Method for Real Variables 325
and
k3 + 6k2 + 17k − 4 ≥ 7k2 + 17k − 4 > 4(k2 − k + 2),
we have
Solution. This inequality is the inequality from the preceding P 3.58 for
1
k= ,
4
but the proof of P 3.58 given in the case k > 1 and using Corollary 2 for δ = 0 fails.
To prove this inequality, we will apply Corollary 2 for Fγ,δ = hγ,δ , where
1 2
δ= = ,
2(k + 5) 21
3−k −44
γ= = .
(k − 1)(k + 5) 63
We have
(
f6 (x, y, z) = 2 yz(k y 2 + z 2 + x 2 )(kz 2 + x 2 + y 2 ) + (k x 2 + y 2 + z 2 ),
A = (k − 1)2 (k + 5),
where
we have
A = 7(−3)3 = −189.
By Corollary 1, we only need to prove the original inequality for y = z = 1. Thus,
we need to show that
1 2 7
+ ≤ ,
x 2 + 9 4x 2 + 5 4x 2 + 6x + 11
which is equivalent to
(x − 1)2 (2x − 7)2 ≥ 0.
The equality holds for x = y = z, and also for 2x = 7 y = 7z (or any cyclic
permutation).
Highest Coefficient Cancellation Method for Real Variables 327
4 3 (4 + 3)2
+ ≥ ,
x2 + y +z
2 2 x y + yz + z x 4(x + y + z 2 ) + 3(x y + yz + z x)
2 2
P = 14(x 2 + y 2 + z 2 ) + x y + yz + z x.
The highest coefficient
+ A of f6 (x, y, z) is equal with the highest coefficient of the
product −45 (4x 2 + y 2 + z 2 ). Since
4x 2 + y 2 + z 2 = 3x 2 + 4(p2 − 2q),
we have
A = −45(3)3 < 0.
By Theorem 1, we only need to prove the original inequality for y = z = 1 and
2x + 1 ≥ 0. Thus, we need to prove that
1 4 45
+ ≥ ,
2x 2 + 1 x 2 + 5 14x 2 + 2x + 29
which is equivalent to
(x − 1)2 (2x + 1)(x + 2) ≥ 0.
The equality holds for x = y = z, and also for −2x = y = z (or any cyclic
permutation).
with equality for x = y = z, and also for −2x = y = z (or any cyclic permutation).
For
(
f6 (x, y, z) = P(x, y, z) (k y 2 +z 2 +x 2 )(kz 2 +x 2 + y 2 )−27(2k+7) (k x 2 + y 2 +z 2 ),
f6 (x, 1, 1)
=P(x, 1, 1)[x 2 + k + 1 + 2(k x 2 + 2)] − 27(2k + 7)(x 2 + k + 1)(k x 2 + 2)
x2 + k + 1
= 2(k − 1)2 (x − 1)2 (2x + 1)[2(k + 5)x + 5k + 16] ≥ 0
for 2x + 1 ≥ 0.
P = 44(x 2 + y 2 + z 2 ) + x y + yz + z x.
The highest coefficient
+ A of f6 (x, y, z) is equal with the highest coefficient of the
product −45 (x 2 + 4 y 2 + 4z 2 ). Since
we have
A = −45(−3)3 = 1215.
Highest Coefficient Cancellation Method for Real Variables 329
Since
f6 (x, y, z)
= (44x 2 + 2x + 89)[4x 2 + 5 + 2(x 2 + 8)] − 45(4x 2 + 5)(x 2 + 8)
4x 2 + 5
= 3(44x 2 + 2x + 89)(2x 2 + 7) − 45(4x 2 + 5)(x 2 + 8)
= 3(x − 1)2 (2x + 1)(14x + 23),
for 2x + 1 ≥ 0. We have
20(x − 1)4 (2x + 1)2
Af−1/2,−2 (x) = ,
3
(x − 1)2 (2x + 1) f (x)
f6 (x, 1, 1) − Af−1/2,−2 (x) = ,
3
where
f (x) = 9(4x 2 + 5)(14x + 23) − 20(x − 1)2 (2x + 1).
Since 4x + 5 > (x − 1)2 , we have
2
f (x) > 9(x − 1)2 (14x + 23) − 20(x − 1)2 (2x + 1) = (x − 1)2 (86x + 187) ≥ 0.
The equality holds for x = y = z, and also for −2x = y = z (or any cyclic
permutation).
Observation 1. Similarly, we can prove the following generalization:
• Let x, y, z be real numbers such that x y + yz + z x ≥ 0. If 0 ≤ k < 1, then
1 27(2k + 7)
≥ .
k x 2 + y 2 + z2 (k + 8)(5k + 5)(x 2 + y 2 + z 2 ) + 2(k − 1)2 (x y + yz + z x)
with equality for x = y = z, and also for −2x = y = z (or any cyclic permutation).
For
(
f6 (x, y, z) = P (k y 2 + z 2 + x 2 )(kz 2 + x 2 + y 2 ) − 27(2k + 7) (k x 2 + y 2 + z 2 ),
f (x) = 27(x 2 + k + 1)[2(k + 5)x + 5k + 16] − 2(2k + 7)(1 − k)(x − 1)2 (2x + 1).
Since
which is equivalent to
6x + 8k + 25 ≥ 0.
Since
y 2 + z 2 = −x 2 + p2 − 2q, −x + 4 y + 4z = −5x + 4p,
Highest Coefficient Cancellation Method for Real Variables 331
that is
A = 9.
Since
f6 (x, 1, 1)
= 21(x 2 + 1) + x(x − 8)(x 2 + 1) − 4(4x + 3)
2(x 2 + 1)
= x 4 − 8x 3 + 22x 2 − 24x + 9 = (x − 1)2 (x − 3)2 ,
f (x) = 225(x 2 + 1) − 2(x − 1)2 > 4(x 2 + 1) − 2(x − 1)2 = 2(x + 1)2 ≥ 0.
The equality holds for x = y = z, and also for x/3 = y = z (or any cyclic
permutation).
x 2 + 3 yz y 2 + 3z x z 2 + 3x y
+ 2 + 2 ≥ 1.
y2 − yz + z 2 z − zx + x 2 x − x y + y2
Since
y 2 − yz + z 2 = −x 2 − yz + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as
where
(
P2 (x, y, z) = (x 2 + 3 yz)(−z 2 − x y)(− y 2 − xz), P3 (x, y, z) = (−x 2 − yz).
Since the original inequality is an equality for (x, y, z) = (0, 1, −1), that means
f6 (0, 1, −1) = 0,
1 h (−2)
γ= + ,
3 12A
1
f6 (x, 1, 1) − Ahγ,δ (x) = (7x 6 + 6x 4 + 156x 3 + 39x 2 − 180x + 188)
56
1
= (x + 2)2 (7x 4 − 28x 3 + 90x 2 − 92x + 47)
56
1
≥ (x + 2)2 (7x 4 − 28x 3 + 74x 2 − 92x + 46)
56
1
= (x + 2)2 [7x 2 (x − 2)2 + 46(x − 1)2 ] ≥ 0.
56
Since
where
(
P2 (x, y, z = x 2 (−2 y 2 − 3z x)(−2z 2 − 3x y), P3 (x, y, z) = (−2x 2 − 3 yz),
that is
Since
f6 (x, 1, 1)
= 2(2x − 1)2 (2x 2 − 3x + 2) + (x − 3)2 − 6(2x 2 − 3x + 2)
2(2x 2 − 3x + 2)
= 16x 4 − 40x 3 + 33x 2 − 10x + 1 = (x − 1)2 (4x − 1)2 ,
The equality holds for x = y = z, and also for 4x = y = z (or any cyclic permu-
tation).
Observation. Similarly, we can prove the following generalization:
2− 4 + 2k
• Let x, y, z be real numbers. If −2 < k ≤ 1 − 5 and m = , then
4
(m y + mz − x)2 3(2m − 1)2
≥ .
y 2 + k yz + z 2 k+2
x = y = z, and also for x/α = y = z (or any cyclic permutation),
with equality for
−k − 2k + 4
where α = .
2
Since
2 + 4 + 2k
• Let x, y, z be real numbers. If −2 < k ≤ 2 and m = , then
4
(m y + mz − x)2 3(2m − 1)2
≥ .
y + k yz + z
2 2 k+2
for x = y = z, and also for x/α = y = z (or any cyclic permutation),
with equality
2k + 4 − k
where α = .
2
336 Vasile Cîrtoaje
From ( (
(x 2 + k x y + y 2 ) = (p2 − 2q + k x y − z 2 ),
it follows that f6 (x, y, z) has the same highest coefficient as
(
4 (k x y − z 2 ) − (2 − k)(x − y)2 ( y − z)2 (z − x)2 ,
f6 (0, y, z) = 4 y 2 z 2 ( y 2 + k yz + z 2 ) − (2 − k) y 2 z 2 ( y − z)2 ,
1 h (−2)
γ= + ,
3 12A
with
h(x) = f6 (x, 1, 1) = 4(k + 2)(x 2 + k x + 1)2 .
We only need to show that f6 (x, 1, 1) ≥ Ahγ,0 (x) for x ∈ . We have
therefore
f6 (x, 1, 1) = Ahγ,0 (x).
A = −1 + 2(3 + 3 + 3) − 1 = 16.
1 h (−2)
γ= + ,
3 12A
with
h(x) = f6 (x, 1, 1) = 4x 4 + 4x 3 + 9x 2 + 4x + 4 = (2x 2 + x + 2)2 .
We only need to show that f6 (x, 1, 1) ≥ Ahγ,0 (x) for x ∈ . We have
therefore
f6 (x, 1, 1) = Ahγ,0 (x).
Solution. Denote the left side of the inequality by f6 (x, y, z). According to (3.3),
the polynomial f6 (x, y, z) has the highest coefficient
2 2(k + 1)3
A = 3 − 3 + (1 + k)3 = > 0.
k k
Since
f6 (0, y, z) = y 6 + z 6 + 2 y 3 z 3 , f6 (0, 1, −1) = 1 + 1 − 2 = 0,
we apply Corollary 2 for Fγ,δ = hγ,δ , where γ is given by (3.19):
1 h (−2)
γ= + ,
3 12A
with
2
h(x) = f6 (x, 1, 1) = x 6 − 3x 2 + 2 + (x 2 + k)(k x + 1)2 .
k
We have
4
h (x) = 6x 5 − 6x +x(k x + 1)2 + 4(x 2 + k)(k x + 1),
k
−4(2k3 + 15k2 + 33k + 2)
h (−2) = ,
k
1 2k3 + 15k2 + 33k + 2 −3k(k + 3)
γ= − = ,
3 6(k + 1)3 2(k + 1)3
Thus, the condition f6 (x, 1, 1) ≥ Ahγ,δ (x) is equivalent to
2
x 6 − 3x 2 + 2 + (x 2 + k)(k x + 1)2 ≥
k
Highest Coefficient Cancellation Method for Real Variables 339
2
2(k + 1)3 3k(k + 3)
≥ x− (x + 2)(2x + 1) + δ(x + 2)3 .
k 2(k + 1)3
k(k + 3)
Setting x = 1, this inequality becomes an equality for δ = . Choosing this
2(k + 1)3
δ, the inequality turns into
that is
A = 2 − 89 + 9 · 27 = 156.
Since
f6 (0, y, z) = 2( y 2 + z 2 )(2 y 2 + z 2 )(2z 2 + y 2 ) − 9 y 2 z 2 ( y − z)2 ,
f6 (0, 1, −1) = 36 − 36 = 0,
we apply Corollary 2 for Fγ,δ = hγ,δ , where γ is given by (3.19):
1 h (−2)
γ= + ,
3 12A
with
h(x) = f6 (x, 1, 1) = 4(x 2 + 1)(x 2 + 3)2 − 89x 2 .
340 Vasile Cîrtoaje
We have
h (x) = 8x(x 2 + 3)3 + 16x(x 2 + 1)(x 2 + 3) − 178x,
1 43 −77
h (−2) = −1548, γ=
− = ,
3 52 156
Thus, the condition f6 (x, 1, 1) ≥ Ahγ,δ (x) is equivalent to
2
77
4(x 2 + 1)(x 2 + 3)2 − 89x 2 ≥ 156 x − (x + 2)(2x + 1) + δ(x + 2)3 .
156
17
Setting x = 1, we can check that this inequality becomes an equality for δ = .
156
17
Choosing δ = , the inequality becomes
156
13x − 1 13 y − 1 13z − 1 3
+ 2 + ≤ .
x 2 + 23 y + 23 z 2 + 23 2
A = 3 · 729.
f6 (0, −1, 1) = 0.
Highest Coefficient Cancellation Method for Real Variables 341
Also, we have
13x 2 + 154x + 167 < 14(2x 2 + 11x + 12) = 14(x + 4)(2x + 3) < 0.
We have
1 h (−2) −5
h (−2) = −1728, + = ,
3 12A 9
2
5
h−5/9,β (x) = x − (x + 2)(2x + 1) + β(x + 2)3 ,
9
2
Ah−5/9,β (x) = 2 9x − 5(x + 2)(2x + 1) + 9β(x + 2)3 ,
f6 (x, 1, 1) − Ah−5/9,β (x) = (x + 2)3 Hβ (x),
Highest Coefficient Cancellation Method for Real Variables 343
where
and for
Eα,β = h−5/9,β , Eγ,δ = h−5/9,δ .
The condition (a), namely f6 (x, 1, 1) ≥ Ah−5/9,β (x) for x < −1/2, can be satisfied
10
only if Hβ (−2) = 0. This yields β = and
81
5
f6 (x, 1, 1) − Ah−5/9,10/81 (x) = (x + 2)4 (41x 2 − 88x + 38) ≥ 0.
81
The condition (b) is satisfied if there is a real δ such that f6 (x, 1, 1) ≥ Ah−5/9,δ (x)
4
for x ≥ −1/2, that is Hδ (x) ≥ 0 for x ≥ −1/2. Choosing δ = , the condition
27
Hδ (x) ≥ 0 is equivalent to
2 16
η= , ξ= ,
5 5
2 2
= , 2 , \ = −∞, ∪ [2, ∞),
5 5
and for
Eα,β = h−5/9,4/27 , Eγ,δ = h−5/9,10/81 .
The condition (a) is satisfied if f6 (x, 1, 1) ≥ Ah−5/9,4/27 (x) for 2/5 < x < 2. As
shown in the first proof, this inequality can be written as
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 in P 3.72 has
the best possible value. Indeed, setting x = 0, y = 1, z = −1 in
we get β ≤ 10. In addition, for β = 10, the best value of the coefficient α of the
product x 2 y 2 z 2 is 108. Setting x = 5t, y = 2t + 1, z = 2t − 1, the inequality
becomes
A(α)t 6 + B(α)t 4 + C(α)t 2 ≥ 0,
where C(α) = 108 − α. The necessary condition C(α) ≥ 0 involves α ≤ 108.
We have
and for
Eα,β = h−2/3,β , Eγ,δ = h−2/3,δ .
The condition (a), namely f6 (x, 1, 1) ≥ Ah−2/3,β (x) for x < −1/2, can be satisfied
2
only if Hβ (−2) = 0. This yields β = and
27
1
f6 (x, 1, 1) − Ah−2/3,2/27 (x) = (x + 2)4 (23x 2 − 88x + 2) ≥ 0
27
The condition (b) is satisfied if there is a real δ such that f6 (x, 1, 1) ≥ Ah−2/3,δ (x)
1
for x ≥ −1/2, that is Hδ (x) ≥ 0 for x ≥ −1/2. Choosing δ = , we get
6
4Hγ (x) = x 3 + 6x 2 − 8x + 8
≥ 2x 2 − 8x + 8 = 2(x − 2)2 ≥ 0.
x 2 + 2 yz y 2 + 2z x z 2 + 2x y 3(x y + yz + z x)
+ 2 + 2 ≥ .
y2 + yz + z 2 z + zx + x 2 x + x y + y2 x 2 + y 2 + z2
Since
y 2 + yz + z 2 = yz − x 2 + p2 − 2q,
f8 (x, y, z) has the same highest polynomial as
(
(x 2 + y 2 + z 2 ) (x 2 + 2 yz)(z x − y 2 )(x y − z 2 ) − 3(x y + yz + z x) ( yz − x 2 ),
A(p, q) = (x 2 + y 2 + z 2 ) · 0 − 3(x y + yz + z x) · 0 = 0.
f8 (x, 1, 1)
= (x 2 + 2)[(x 2 + 2)(x 2 + x + 1) + 6(2x + 1)] − 9(2x + 1)(x 2 + x + 1)
x2 + x + 1
= x 6 + x 5 + 5x 4 − 2x 3 − 13x 2 + x + 7
= (x 2 − 1)2 (x 2 + x + 7) ≥ 0.
The equality holds for x = y = z, and for −x = y = z (or any cyclic permuta-
tion).
x 2 − 2 yz y 2 − 2z x z 2 − 2x y 3(x y + yz + z x)
+ + + ≥ 0.
y 2 − yz + z 2 z 2 − z x + x 2 x 2 − x y + y 2 x 2 + y 2 + z2
Since
y 2 − yz + z 2 = − yz − x 2 + p2 − 2q,
f8 (x, y, z) has the same highest polynomial as
(
(x 2 + y 2 + z 2 ) (x 2 − 2 yz)(z x + y 2 )(x y + z 2 ) − 3(x y + yz + z x) ( yz + x 2 ),
The equality holds for x = y = z, and for −x = y = z (or any cyclic permuta-
tion).
x2 y2 z2 (x + y + z)2
+ + ≥ .
y 2 − yz + z 2 z 2 − z x + x 2 x 2 − x y + y 2 x 2 + y 2 + z2
(Vasile C., 2014)
Solution. Write the inequality as f8 (x, y, z) ≥ 0, where
f8 (x, y, z) =(x 2 + y 2 + z 2 ) x 2 (z 2 − z x + x 2 )(x 2 − x y + y 2 )
(
− (x + y + z)2 ( y 2 − yz + z 2 ).
Since
y 2 − yz + z 2 = − yz − x 2 + p2 − 2q,
f8 (x, y, z) has the same highest polynomial as
(
(x 2 + y 2 + z 2 ) x 2 (z x + y 2 )(x y + z 2 ) + (x + y + z)2 ( yz + x 2 ),
We have
f8 (x, 1, 1)
= (x 2 + 2)[x 2 (x 2 − x + 1) + 2] − (x + 2)2 (x 2 − x + 1)
x2 − x + 1
= x 6 − x 5 + 2x 4 − 5x 3 + 3x 2
= x 2 (x − 1)2 (x 2 + x + 3) ≥ 0.
The equality holds for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Since
y 2 z 2 ( y + z)2 = y 2 z 2 (p − x)2 = p2 y 2 z 2 − 2pqr + 3r 2 ,
A(p, q) = 3(x 2 + y 2 + z 2 )(3 − 2) − 10(x + y + z)2 = 3(p2 − 2q) − 10p2 = −7p2 − 6q.
Highest Coefficient Cancellation Method for Real Variables 349
We have
Case 1: −7p2 − 6q ≤ 0. Apply Theorem 3. Since f8 (x, 1, 1) ≥ 0 for all real x, the
conclusion follows.
Eα,β = hα,β .
Thus, we need to show that there exist two real numbers α and β such that
Since
h(−2) = 72 − 18(−2)2 = 0,
the condition h (−2) = 0 is necessary to have h(x) ≥ 0 in the vicinity of −2. This
condition involves α = −1. For this α, we can check that h (−2) = 0. Thus, to
have h(x) ≥ 0 in the vicinity of −2, the condition h (−2) = 0 is necessary. This
condition involves β = 2/3. For these values of α and β, we need to show that
2
2
f8 (x, 1, 1) + (7x 2 + 40x + 34) x − (x + 2)(2x + 1) + (x + 2)3 ≥0
3
14x 4 +52x 3 +117x 2 +154x +113 = (x 2 +1)2 +13x 2 (x +2)2 +7(9x 2 +22x +16) > 0.
that is
We have
A(p, q) = (1 + k)[3p2 + 4(p2 − 3q)] ≥ 0,
A(x + 2, 2x + 1) = (1 + k)[7(x + 2)2 − 12(2x + 1)] = (1 + k)(7x 2 + 4x + 16),
f8 (x, 1, 1)
= 2(x 2 + 2)[(x 2 + k)(x 2 + 1) + 4(k x + 1)] − 2(1 + k)(x + 2)2 (x 2 + 1),
x2 + 1
f8 (x, 1, 1) = 2(x 2 + 1)(x − 1)2 (x 4 + 2x 3 + 5x 2 + 4x + 4 − 2k).
49
For k = , we get
32
81
A(x + 2, 2x + 1) = (7x 2 + 4x + 16),
32
1
f8 (x, 1, 1) = (x 2 + 1)(x − 1)2 (2x + 1)2 (4x 2 + 4x + 15).
8
Since A(p, q) ≥ 0 for all real x, y, z, f8 (1, 1, 1) = 0 and f8 (−1/2, 1, 1) = 0, we apply
Corollary 4 for
4
E−1/2,−2 (x) = f−1/2,−2 (x) = (x − 1)4 (2x + 1)2 .
729
Highest Coefficient Cancellation Method for Real Variables 351
The equality holds for x = y = z, and also for −2x = y = z (or any cyclic
permutation).
x 2 + 4 yz y 2 + 4z x z 2 + 4x y 15(x y + yz + z x)
(a) + 2 + 2 ≥ ;
y 2 + z2 z + x2 x + y2 2(x 2 + y 2 + z 2 )
2x 2 + 9 yz 2 y 2 + 9z x 2z 2 + 9x y 33(x y + yz + z x)
(b) + 2 + ≥ .
y 2 + z2 z + x2 x2 + y2 2(x 2 + y 2 + z 2 )
(Vasile C., 2014)
that is
We have
A(p, q) = 3(1 + k)[p2 + (p2 − 3q)] ≥ 0,
A(x + 2, 2x + 1) = 3(1 + k)[2(x + 2)2 − 3(2x + 1)] = 3(1 + k)(2x 2 + 2x + 5),
f8 (x, 1, 1)
= 2(x 2 + 2)[(x 2 + k)(x 2 + 1) + 4(k x + 1)] − 6(1 + k)(2x + 1)(x 2 + 1),
x2 + 1
f8 (x, 1, 1) = 2(x 2 + 1)(x − 1)2 [x 4 + 2x 3 + (k + 6)x 2 + 4x + 5 − k].
Since A(p, q) ≥ 0 for all real x, y, z, we apply Corollary 4 for
1 2
E0,−2 (x) = f0,−2 (x) = x (x − 1)4 .
81
Thus, we need to show that
for x ∈ . Since
(1 + k)(2x 2 + 2x + 5)x 2 (x − 1)4
A(x + 2, 2x + 1) f0,−2 (x) = ,
27
the inequality f8 (x, 1, 1) − A(x + 2, 2x + 1) f0,−2 (x) ≥ 0 is true if
g(x) = (25 − 2k)x 4 + (52 − 2k)x 3 + (22k + 157)x 2 + 108x + 27(5 − k).
x2 y2 z2 4(x y + yz + z x)
+ + ≥ .
( y − z) 2 (z − x) 2 (x − y)2 x 2 + y 2 + z2
Since
(x − y)(x − z) = x 2 + 2 yz − q,
f8 (x, y, z) has the same highest polynomial as
(p2 − 2q) x 2 (x 2 + 2 yz)2 − 4q(x − y)2 ( y − z)2 (z − x)2 ,
We have
f8 (x, 1, 1) = (x 2 + 2)x 2 (x − 1)4 ≥ 0.
4x 2 (x − 1)4 (x − γ)2
Eγ,δ (x) = fγ,0 (x) = .
9(4 − γ)2 (x + 2)2
Thus, we need to show that there exist a real number α such that
for x ∈ . Since
A(x + 2, 2x + 1) = 27(x 2 + 8x + 6),
12x 2 (x − 1)4 (x 2 + 8x + 6)(x − γ)2
A(x + 2, 2x + 1) fγ,0 (x) = ,
(4 − γ)2 (x + 2)2
x 2 (x − 1)4 h(x)
f8 (x, 1, 1) − A(x + 2, 2x + 1) fγ,0 (x) = ,
(4 − γ)2 (x + 2)2
h(x) = (4 − γ)2 (x + 2)2 (x 2 + 2) − 12(x 2 + 8x + 6)(x − γ)2 ,
354 Vasile Cîrtoaje
we need to show that h(x) ≥ 0 for x ∈ . Since h(4) = 0, the condition h (4) = 0
is necessary to have h(x) ≥ 0 in the vicinity of 4. This implies α = −2/13 and
x2 y2 z2 (x + y + z)2
+ + ≥ 2 .
( y − z) 2 (z − x) 2 (x − y) 2 x + y 2 + z2
which is equivalent to
x 2 + y 2 + z 2 ≤ 2(x y + yz + z x),
p2 ≤ 4q.
Consider next that
p2 ≥ 4q,
and write the inequality as f8 (x, y, z) ≥ 0, where
f8 (x, y, z) = (x 2 + y 2 +z 2 ) x 2 (x − y)2 (x −z)2 −(x + y +z)2 (x − y)2 ( y −z)2 (z− x)2 .
Since
(x − y)(x − z) = x 2 + 2 yz − q,
f8 (x, y, z) has the same highest polynomial as
(x 2 + y 2 + z 2 ) x 2 (x 2 + 2 yz)2 − (x + y + z)2 (x − y)2 ( y − z)2 (z − x)2 ,
Highest Coefficient Cancellation Method for Real Variables 355
Since
A(p, q) = 18[2p2 + (p2 − 3q)] ≥ 0,
A(x + 2, 2x + 1) = 54(x 2 + 2x + 3),
f8 (x, 1, 1) = (x 2 + 2)x 2 (x − 1)4 ,
we apply Corollary 4 for p2 ≥ 4q (see Remark 6) and
1 2
Eγ,δ (x) = f0,−2 (x) = x (x − 1)4 .
81
Thus, we need to show that
Since
2(x 2 + 2x + 3)x 2 (x − 1)4
A(x + 2, 2x + 1) f0,−2 (x) = ,
3
we have
x 2 (x − 1)4 x(x − 4)
f8 (x, 1, 1) − A(x + 2, 2x + 1) f0,−2 (x) = ≥ 0.
3
2x y 2 yz 2z x 4(x y + yz + z x)
+ + +1≥ .
x 2 + y 2 y 2 + z2 z2 + x 2 x 2 + y 2 + z2
Since
y 2 + z 2 = −x 2 + p2 − 2q,
356 Vasile Cîrtoaje
the polynomial of degree eight f8 (x, y, z) has the same highest polynomial A(p, q)
as
2(p2 − 2q) yz(−z 2 )(− y 2 ) + (p2 − 6q)(−z 2 )(−x 2 )(− y 2 ),
that is
A(p, q) = 6(p2 − 2q) − (p2 − 6q) = 5p2 − 6q = 2(p2 − 3q) + 3p2 > 0,
A(x + 2, 2x + 1) = 5x 2 + 8x + 14 > 0.
Since
f8 (x, 1, 1) = 2(x 2 + 2) (x 2 + 1)2 + 4x(x 2 + 1) + 2(x 2 − 8x − 2)(x 2 + 1)2
= 4x 2 (x 2 + 1)(x − 1)2 ,
and
4(x − 1)4 4(x − 1)4 (5x 2 + 8x + 14)
f∞,−2 (x) = , A(x + 2, 2x + 1) f∞,−2 (x) = ,
81 81
4(x − 1)2 g(x)
f8 (x, 1, 1) − A(x + 2, 2x + 1) f∞,−2 (x) = ,
81
where
−1
The condition (a), namely f8 (x, 1, 1) − A(x + 2, 2x + 1) f0,∞ (x) ≥ 0 for x > ,
2
is satisfied because
−1
The condition (b), namely f8 (x, 1, 1) ≥ A(x + 2, 2x + 1) f∞,−2 (x) for x ≤ , is
2
satisfied because
The equality holds for x = y = z, and also for x = 0 and y = z (or any cyclic
permutation).
358 Vasile Cîrtoaje
Chapter 4
359
360 Vasile Cîrtoaje
where A is the highest coefficient of f6 (x, y, z) (see section Theoretical Basis from
Chapter 3), and g1 (p, q) and g2 (p, q) are polynomial functions of the form
are, respectively,
3, −3, −3, 3, 3, −3.
has the highest coefficients P1 (1, 1, 1), P2 (1, 1, 1), P3 (1, 1, 1) and −27, respectively.
Based on Lemma above, Theorem 1 bellow gives for A ≤ 0 the necessary and
sufficient conditions to have f6 (x, y, z) ≥ 0 for all nonnegative real numbers x, y, z.
Theorem 1 can be extended in the following form (see P 3.76 in Volume 1, page
173):
Theorem 1 . Let f6 (x, y, z) be a sixth degree symmetric homogeneous polynomial
having the highest coefficient A ≤ 0.
(a) The inequality f6 (x, y, z) ≥ 0 holds for all nonnegative real numbers x, y, z
satisfying p2 ≤ 4q if and only if f6 (x, 1, 1) ≥ 0 for x ∈ [0, 4] ;
(b) The inequality f6 (x, y, z) ≥ 0 holds for all nonnegative real numbers x, y, z
satisfying p2 ≥ 4q if and only if f6 (x, 1, 1) ≥ 0 for x ≥ 4 , and f6 (0, y, z) ≥ 0 for
y, z ≥ 0.
Highest Coefficient Cancellation Method for Nonnegative Variables 361
∗∗∗
f6 (x, y, z) ≥ 0,
holds for all nonnegative real numbers x, y, z if and only if it holds for y = z = 1
and for x = 0 (see Theorem 1);
3) treating successively the cases p2 < 4q and p2 ≥ 4q.
∗∗∗
2
(2x + 1)2
gα,β (x) = x + α(x + 2)(2x + 1) + β , (4.4)
x +2
y 2z2 2
ĝα,β ( y, z) = α( y + z)2 + β yz , (4.5)
( y + z)2
362 Vasile Cîrtoaje
2
hα,β (x) = x + α(x + 2)(2x + 1) + β(x + 2)3 , (4.6)
2
ĥα,β ( y, z) = ( y + z)2 α yz + β( y + z)2 . (4.7)
The functions fα,β (x) and fˆα,β ( y, z) have been derived by setting respectively
y = z = 1 and x = 0 in the associated function
2
b1 2 c1 q2
f¯6 (x, y, z) = r − pq + p3 + · (4.8)
a1 a1 a1 p
with
The functions gα,β (x) and ĝα,β ( y, z) have been derived by setting respectively
y = z = 1 and x = 0 in the associated function
2
q2
f¯6 (x, y, z) = r + αpq + β .
p
The functions hα,β (x) and ĥα,β ( y, z) have been derived by setting respectively
y = z = 1 and x = 0 in the associated function
2
f¯6 (x, y, z) = r + αpq + β p3 .
∗∗∗
With regard to the functions fα,β (x) and fˆα,β (x), we get the following expressions
for β = −2, β = −1, β = 0, β = 1 and β → ∞:
In particular,
With regard to the functions gα,β (x) and ĝα,β (x), we get the following particular
expressions:
With regard to the functions hα,β (x) and ĥα,β (x), we have
Notice that the relative degree of the rational functions fα,β (x), fα,∞ (x) and
f∞,∞ (x) are six, four and two, respectively. Also, gα,β (x) and g0,0 (x) have the rel-
ative degree equal to four and two, respectively, while hα,β (x), hα,0 (x) and h0,0 (x)
have the relative degree equal to six, four and two, respectively.
∗∗∗
Proof. Let
2
q2
E1 (x, y, z) = r + A1 pq + A2 p3 + A3
p
and 2
q2
F1 (x, y, z) = r + B1 pq + B2 p3 + B3
p
be the functions associated to Eα,β (x) and Fγ,δ (x), respectively; this means that
and
F2 (x, y, z) = f6 (x, y, z) − AF1 (x, y, z).
Since AE1 (x, y, z) ≥ 0 and AF1 (x, y, z) ≥ 0, the inequality f6 (x, y, z) ≥ 0 holds for
all nonnegative real x, y, z if
(a) E2 (x, y, z) ≥ 0 for p2 ≤ 4q;
(b) F2 (x, y, z) ≥ 0 for p2 ≥ 4q.
According to Theorem 1 , since E2 (x, y, z) and F2 (x, y, z) has the highest coefficient
zero, these conditions are satisfied if and only if
(a) E2 (x, 1, 1) ≥ 0 for (x + 2)2 ≤ 4(2x + 1);
(b) F2 (x, 1, 1) ≥ 0 for (x +2)2 ≥ 4(2x +1), and F2 (0, y, z) ≥ 0 for ( y +z)2 ≥ 4 yz.
Since these conditions are equivalent to the condition (a), (b) and (c) in Theorem
2, the proof is completed.
Remark 1. The function f¯6 (x, y, z) associated to fα,β (x) (given by (4.8)) is zero for
If β = −2, then the function associated to fα,−2 (x) has the expression
2
α+8 2
f¯6 (x, y, z) = r − pq + p3 , (4.9)
9α + 18 9α + 18
In addition, if α → ∞, then
2
1
f¯6 (x, y, z) = r − pq
9
is zero for
(x, y, z) = (1, 1, 1), (x, y, z) = (1, 0, 0).
If β → ∞, then
2
1 2α + 4 q2
f¯6 (x, y, z) = r + pq − ·
6α + 3 6α + 3 p
is zero for
is zero for
(x, y, z) = (1, 1, 1), (x, y, z) = (1, 0, 0).
Remark 2. If
f6 (x, 1, 1) = Afγ,−2 (x)
for all x ∈ , then there is k ≥ 0 such that the following identity holds:
k = 0,
Remark 3. Theorem 2 is also valid for the case where the parameters α and β of
the function fα,β (x) are conjugate complex numbers. For example, if k > 0 and
α = −k, β = − −k,
(x − 1)4 (x 2 + k)2
f
−k,−
−k = . (4.11)
9(k − 2)2 (x + 2)2
∗∗∗
Proof. From
we get
9pq − 2p3 − 2(p2 − 3q) p2 − 3q
r≥ .
27
Define the nonnegative function
9pq − 2p3 − 2(p2 − 3q) p2 − 3q
E(x, y, z) = r r − ,
27
E(0, y, z) = 0,
⎧
⎨ 0, 0≤ x ≤1
E(x, 1, 1) = .
⎩ 4x(x − 1) , x ≥ 1
3
27
Let us denote
E1 (x, y, z) = f6 (x, y, z) − AE(x, y, z).
368 Vasile Cîrtoaje
Since AE(x, y, z) ≥ 0, the inequality f6 (x, y, z) ≥ 0 holds for all nonnegative real
x, y, z if E1 (x, y, z) ≥ 0 for x, y, z ≥ 0. According to Theorem 1, because E1 (x, y, z)
has the highest coefficient zero, the inequality E1 (x, y, z) ≥ 0 holds for all x, y, z ≥ 0
if and only if E1 (x, 1, 1) ≥ 0 and E1 (0, y, z) ≥ 0 for all nonnegative numbers x, y, z.
These conditions are equivalent to the condition (a), (b) and (c) in Theorem 3.
Thus, the proof is completed.
Remark 4. Theorem 2 and Corollary 1 are also valid by replacing the condition (a)
with the following two conditions:
(a1 ) f6 (x, 1, 1) ≥ 0 for 0 ≤ x ≤ 1 ;
4Ax(x − 1)3
(a2 ) f6 (x, 1, 1) ≥ for 1 ≤ x ≤ 4.
27
A(p, q) = μ1 p2 + μ2 q, μ1 , μ2 ∈ .
Theorems 1, 1 , 2 and 3 can be extended to these polynomials as follows:
Theorem 4. Let f (x, y, z) be a symmetric homogeneous polynomial of degree seven
or eight which has the highest polynomial A(p, q). The inequality f (x, y, z) ≥ 0
holds for all nonnegative real numbers x, y, z satisfying A(p, q) ≤ 0 if and only if
f (x, 1, 1) ≥ 0 and f (0, y, z) ≥ 0 for all nonnegative real numbers x, y, z such that
A(x + 2, 2x + 1) ≤ 0 and A( y + z, yz) ≤ 0.
4.2 Applications
4.4. If x, y, z are nonnegative real numbers, no two of which are zero, then
1 1 1 9
+ + ≥ 2 .
4x 2 + yz 4 y 2 + z x 4z 2 + x y (x + 4 yz) + ( y 2 + 4z x) + (z 2 + 4x y)
2x 2 − yz 2 y2 − z x 2z 2 − x y 1
+ + 2 ≤ .
4x + y + z
2 2 2 4y + z + x
2 2 2 4z + x 2 + y 2 2
21x 2 + 4 yz 21 y 2 + 4z x 21z 2 + 4x y
+ + ≥ 15.
x 2 + 2 y 2 + 2z 2 y 2 + 2z 2 + 2x 2 z 2 + 2x 2 + 2 y 2
11
4.9. Let x, y, z be nonnegative real numbers. If k ≤ , then
2
x(2x + y +z)(x − y)(x −z)(x −k y)(x −kz)+(7k−13)(x − y)2 ( y −z)2 (z− x)2 ≥ 0.
Then,
αk (x − y)2 ( y − z)2 (z − x)2
(x − y)(x − z)(x − k y)(x − kz) ≥ .
x y + yz + z x
4
Then,
αk (x − y)2 ( y − z)2 (z − x)2
x(x − y)(x − z)(x − k y)(x − kz) + ≥ 0.
x + y +z
4.27. If x, y, z are nonnegative real numbers, no two of which are zero, then
2 2 2 45
+ + ≥ .
x 2 + y 2 y 2 + z2 z2 + x 2 4(x 2 + y 2 + z 2 ) + x y + yz + z x
4.28. If x, y, z are nonnegative real numbers, no two of which are zero, then
1 18
≥ .
2 y 2 + yz + 2z 2 5(x 2 + y 2 + z 2 + x y + yz + z x)
4.29. If x, y, z are nonnegative real numbers, no two of which are zero, then
x − (2 +
2)( y + z)
9(3 + 2 2)
+ ≥ 0.
( y + z)2 4(x + y + z)
4.30. If x, y, z are nonnegative real numbers, no two of which are zero, then
6x − y − z 6 y − z − x 6z − x − y 18
+ + ≥ .
y 2 + z2 z2 + x 2 x2 + y2 x + y +z
Highest Coefficient Cancellation Method for Nonnegative Variables 375
4.31. If x, y, z are nonnegative real numbers, no two of which are zero, then
2x − 3 y − 3z 6
+ ≥ 0.
y 2 + 4 yz + z 2 x + y + z
4.32. If x, y, z are nonnegative real numbers, no two of which are zero, then
7x + 4 y + 4z 27
≥ .
4x 2 + yz x + y +z
4.36. If x, y, z are nonnegative real numbers, no two of which are zero, then
2x 2 + yz 9 31(x − y)2 ( y − z)2 (z − x)2
≥ + .
y +z
2 2 2 2(x 2 + y 2 )( y 2 + z 2 )(z 2 + x 2 )
4.37. If x, y, z are nonnegative real numbers, no two of which are zero, then
2x 2 − yz 9(x − y)2 ( y − z)2 (z − x)2
≥3+ 2 .
y 2 − yz + z 2 (x − x y + y 2 )( y 2 − yz + z 2 )(z 2 − z x + x 2 )
4.38. If x, y, z are nonnegative real numbers, no two of which are zero, then
x y − yz + z x 3 5(x − y)2 ( y − z)2 (z − x)2
≥ + .
y +z
2 2 2 2(x 2 + y 2 )( y 2 + z 2 )(z 2 + x 2 )
376 Vasile Cîrtoaje
4.39. If x, y, z are nonnegative real numbers, no two of which are zero, then
x y − 2 yz + z x 3(x − y)2 ( y − z)2 (z − x)2
≥ 2 .
y − yz + z
2 2 (x − x y + y 2 )( y 2 − yz + z 2 )(z 2 − z x + x 2 )
4.40. If x, y, z are nonnegative real numbers, no two of which are zero, then
x + 3 y + 3z 7(x + y + z)
≥ .
( y + 2z)(2 y + z) 3(x y + yz + z x)
4.41. If x, y, z are nonnegative real numbers, no two of which are zero, then
9x − 5 y − 5z 3(x + y + z)
+ ≥ 0.
2 y 2 − 3 yz + 2z 2 x y + yz + z x
4.42. If x, y, z are nonnegative real numbers, no two of which are zero, then
3x − y − z 3(x + y + z)
≥ .
y 2 + z2 2(x y + yz + z x)
1 − 17 1 + 17
k ∈ [a, b], a= ≈ −1.56155, b = ≈ 2.56155 ,
2 2
then (3 − k)x + (k − 1)( y + z) 3(k + 1) x + y +z
≥ · .
y 2 + k yz + z 2 k+2 x y + yz + z x
4.44. If x, y, z are nonnegative real numbers, no two of which are zero, then
x + 13 y + 13z 27(x + y + z)
≥ .
y 2 + 4 yz + z 2 2(x y + yz + z x)
4.45. If x, y, z are nonnegative real numbers, no two of which are zero, then
−x + y + z x + y +z
≥ .
2x 2 + yz x y + yz + z x
Highest Coefficient Cancellation Method for Nonnegative Variables 377
4.46. If x, y, z are nonnegative real numbers, no two of which are zero, then
11x − 3 y − 3z 3(x + y + z)
≤ .
2x 2 + 3 yz x y + yz + z x
4.47. If x, y, z are nonnegative real numbers, no two of which are zero, then
1 1 1 1 2
+ + ≥ + .
2x 2 + yz 2 y 2 + z x 2z 2 + x y x y + yz + z x x 2 + y 2 + z 2
4.48. If x, y, z are nonnegative real numbers, no two of which are zero, then
x( y + z) y(z + x) z(x + y) x 2 + y 2 + z2
+ 2 + 2 ≤ .
x + 5 yz y + 5z x z + 5x y
2 x y + yz + z x
4.49. If x, y, z are nonnegative real numbers, no two of which are zero, then
x( y + z) y(z + x) z(x + y) 15(x y + yz + z x)
+ 2 + 2 +2≥ .
x 2 + yz y + zx z +xy (x + y + z)2
4.50. If x, y, z are nonnegative real numbers, no two of which are zero, then
x( y + z) y(z + x) z(x + y) x y + yz + z x
+ + ≥1+ 2 .
x 2 + 2 yz y 2 + 2z x z 2 + 2x y x + y 2 + z2
4.54. If x, y, z are nonnegative real numbers, no two of which are zero, then
2 2 2 8 1
+ + ≥ 2 + .
x 2 + y 2 y 2 + z2 z2 + x 2 x + y 2 + z 2 x y + yz + z x
4.3 Solutions
Since
x( y + z) = x(p − x),
(x − y)(x − z) = x 2 + 2 yz − q,
(x − 3 y)(x − 3z) = x 2 + 12 yz − 3q,
f6 (x, y, z) has the same highest coefficient A as
P1 (x, y, z) = −x 2 (x 2 + 2 yz)(x 2 + 12 yz),
that is
A = P1 (1, 1, 1) = −3(1 + 2)(1 + 12) < 0.
By Theorem 1, we only need to show that f6 (x, 1, 1) ≥ 0 and f6 (0, y, z) ≥ 0 for
x, y, z ≥ 0. We have
The equality holds for x = y = z, for x/3 = y = z (or any cyclic permutation),
for x = 0 and y = z (or any cyclic permutation), and for y = z = 0 (or any cyclic
permutation).
Observation. Similarly, we can prove the following generalization:
• Let x, y, z be nonnegative real numbers. If
then
x( y + z)(x − y)(x − z)(x − k y)(x − kz) + (k − 3)(x − y)2 ( y − z)2 (z − x)2 ≥ 0,
with equality
for x = y = z, for x/k = y = z (or any cyclic permutation) if k ≥
−2 + 2 3, for x = 0 and y = z (or any cyclic permutation), and for y = z = 0 (or
any cyclic permutation).
380 Vasile Cîrtoaje
We have
that is
A1 = P1 (1, 1, 1) = 3(1 + 2)(147) = 1323.
Therefore, f6 (x, y, z) has the highest coefficient
The equality holds for x = y = z, for 2x = 11 y = 11z (or any cyclic permuta-
tion), and for x = 0 and y = z (or any cyclic permutation).
Observation. Similarly, we can prove the following generalization:
• Let x, y, z be nonnegative real numbers. If
4k2 + 12k + 17 11
αk = , k≥ ,
8 2
then
x(2x + y + z)(x − y)(x − z)(x − k y)(x − kz) + αk (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
with equality for x = y = z, for x/k = y = z (or any cyclic permutation), for x = 0
y z 2k − 3
and y = z (or any cyclic permutation), and for x = 0 and + = (or any
z y 4
cyclic permutation).
Denoting the left side of the inequality by f6 (x, y, z), we have
4k2 + 12k + 17
A = 3(1 + 2)(1 + k + k2 ) + (−27)
8
−9
= (4k2 + 28k + 387) ≤ 0,
8
f6 (x, 1, 1) = 2x(x + 1)(x − 1)2 (x − k)2 ≥ 0,
1
f6 (0, y, z) = ( y − z)2 [4 y 2 + 4z 2 − (2k − 3) yz]2 ≥ 0.
8
that is
A1 = P1 (1, 1, 1) = 3(1 + 2)(1 + 12) = 117.
Therefore, f6 (x, y, z) has the highest coefficient
The equality holds for x = y = z, for x/3 = y = z (or any cyclic permutation),
and for x = 0 and y = z (or any cyclic permutation).
Observation. Similarly, we can prove the following generalization:
• Let x, y, z be nonnegative real numbers. If
11
10 − 2 15 ≤ k ≤ ,
then 2
x(2x + y +z)(x − y)(x −z)(x −k y)(x −kz)+(7k−13)(x − y)2 ( y −z)2 (z− x)2 ≥ 0,
with equality for x = y = z, for x/k = y = z (or any cyclic permutation), and for
x = 0 and y = z (or any cyclic permutation).
We have
P 4.4. If x, y, z are nonnegative real numbers, no two of which are zero, then
1 1 1 9
+ + ≥ 2 .
4x 2 + yz 4 y 2 + z x 4z 2 + x y (x + 4 yz) + ( y 2 + 4z x) + (z 2 + 4x y)
(Vasile C., 2008)
Highest Coefficient Cancellation Method for Nonnegative Variables 383
The highest coefficient A of f6 (x, y, z) is equal to the highest coefficient of the prod-
uct (
P3 (x, y, z) = −9 (4x 2 + yz),
that is
A = P3 (1, 1, 1) = −1125.
By Theorem 1, we only need to prove the original inequality for y = z = 1, and for
x = 0. Thus, we need to show that
1 2 9
+ ≥ 2
4x 2 + 1 x + 4 x + 8x + 6
and
1 1 1 9
+ + ≥ 2 .
yz 4 y 2 4z 2 y + z 2 + 4 yz
These inequalities are respectively equivalent to
and
( y − z)2 ( y 2 + z 2 + 10 yz) ≥ 0.
The equality holds for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Observation. Similarly, we can prove the following generalization:
• Let x, y, z be nonnegative real numbers, no two of which are zero. If
1 ≤ k ≤ 3 + 5,
then
1 9(k + 2)
≥ ,
k x 2 + yz (−k2 + 6k − 2)(x 2 + y 2 + z 2 ) + (2k2 − 3k + 4)(x y + yz + z x)
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
2x 2 − yz 2 y2 − z x 2z 2 − x y 1
+ + ≤ .
4x 2 + y 2 + z 2 4 y 2 + z 2 + x 2 4z 2 + x 2 + y 2 2
(Vasile C., 2008)
384 Vasile Cîrtoaje
Since
4x 2 + y 2 + z 2 = 3x 2 + p2 − 2q,
the highest coefficient A of f6 (x, y, z) is equal to the highest coefficient of
(
(3x 2 ) − 2P2 (x, y, z),
where
P2 (x, y, z = (2x 2 − yz)(3 y 2 )(3z 2 ),
that is
A = 27 − 2P2 (1, 1, 1) = 27 − 54 = −27.
By Theorem 1, we only need to prove the original inequality for y = z = 1, and for
x = 0. Thus, we need to show that
2x 2 − 1 2(2 − x) 1
+ 2 ≤
2(2x 2 + 1) x +5 2
and
− yz 2 y2 2z 2 1
+ + ≤ .
y 2 + z 2 4 y 2 + z 2 4z 2 + y 2 2
These inequalities are respectively equivalent to
(x − 1)2 (4x + 1) ≥ 0
and
− yz ( y 2 + z 2 )2 + 6 y 2 z 2 1
+2 ≤ .
y +z
2 2 4( y 2 + z 2 )2 + 9 y 2 z 2 2
For yz = 0, the last inequality is an equality. For yz = 0, using the substitution
y 2 + z2
t= , t ≥ 2,
yz
The equality holds for x = y = z, and for y = z = 0 (or any cyclic permutation).
21x 2 + 4 yz 21 y 2 + 4z x 21z 2 + 4x y
+ + ≥ 15.
x 2 + 2 y 2 + 2z 2 y 2 + 2z 2 + 2x 2 z 2 + 2x 2 + 2 y 2
Since
x 2 + 2 y 2 + 2z 2 = −x 2 + 2(p2 − 2q),
f6 (x, y, z) has the same highest coefficient A as
(
P2 (x, y, z) − 15 (−x 2 ),
where
P2 (x, y, z) = (21x 2 + 4 yz)(− y 2 )(−z 2 ),
that is
A = P2 (1, 1, 1) + 15 = 75 + 15 = 90.
Since
The condition (c) in Theorem 3 is satisfied if the original inequality holds for
x = 0. Thus, we need to show that
2 yz ( y 2 + z 2 )2 − y 2 z 2
+ 42 ≥ 15.
y 2 + z2 2( y 2 + z 2 )2 + y 2 z 2
2 42(t 2 − 1)
+ ≥ 15,
t 2t 2 + 1
12t 3 + 4t 2 − 57t + 2 ≥ 0,
(t − 2)(12t 2 + 28t − 1) ≥ 0.
The equality holds for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
(1+k)(4−k) 2
x + 2 yz 3(3 − k)
1−k
≥ ,
k x 2 + y 2 + z2 1−k
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation). If
k = 2, then the equality holds also for y = z = 0 (or any cyclic permutation).
x y − yz + z x yz − z x + x y z x − x y + yz 3
+ + ≤ .
x 2 + 3 y 2 + 3z 2 y 2 + 3z 2 + 3x 2 z 2 + 3x 2 + 3 y 2 7
Since
that is
A = −24 + 168 = 144.
Since
f (x) = 225(3x 2 +4)−16(x −1)2 > 16(3x 2 +4)−16(x −1)2 = 16(2x 2 +2x +3) > 0.
The condition (c) of Corollary 1 is satisfied if the original inequality holds for
x = 0. Thus, we need to show that
− yz yz yz 3
+ + ≤ ,
3( y 2 + z 2 ) y 2 + 3z 2 z 2 + 3 y 2 7
388 Vasile Cîrtoaje
− yz 4 yz( y 2 + z 2 ) 3
+ ≤ .
3( y 2 + z 2 ) 3( y 2 + z 2 )2 + 4 y 2 z 2 7
y 2 + z2
t= , t ≥ 2,
yz
−1 4t 3
+ 2 ≤ ,
3t 3t + 4 7
9t 2 − 4 3
≤ .
3t(3t 2 + 4 7
It suffices to show that
9t 2 3
≤ ,
3t(3t 2 + 4) 8
which reduces to
(t − 2)(3t − 2) ≥ 0.
3x
The equality holds for x = y = z, and for = y = z (or any cyclic permuta-
4
tion).
3k(k − 1) 3k2 − 6k + 1
α= , β= ,
3k2 − 1 3k2 − 1
1
where k > 0, k =
. Then,
3
αx( y + z) − β yz 3(2α − β)
≤ ,
3k2 x 2 + y 2 + z 2 3k2 + 2
with equality for x = y = z, and for x/γ = y = z (or any cyclic permutation), where
2
γ=k+ − 1.
3k
Highest Coefficient Cancellation Method for Nonnegative Variables 389
that is
A1 = P1 (1, 1, 1) = 3(1 + 2) = 9.
Therefore, f6 (x, y, z) has the highest coefficient
A = 9 − 13(−27) = 360.
Since
f6 (x, 1, 1) = 2x 3 (x + 1)(x − 1)2 ,
we apply Corollary 1 for
x 4 (x − 1)4
Eα,β (x) = f0,0 (x) = .
36(x + 2)2
The condition (a) of Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Af0,0 (x) for x ∈ [0, 4].
We have
10x 4 (x − 1)4
f6 (x, 1, 1) − Af0,0 (x) = 2x 3 (x + 1)(x − 1)2 −
(x + 2)2
2x (x − 1) (4 − x)(4x + x + 1)
3 2 2
= ≥ 0.
(x + 2)2
The equality holds for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
11
P 4.9. Let x, y, z be nonnegative real numbers. If k ≤ , then
2
x(2x + y +z)(x − y)(x −z)(x −k y)(x −kz)+(7k−13)(x − y)2 ( y −z)2 (z− x)2 ≥ 0.
that is
A1 = P1 (1, 1, 1) = 9(k2 + k + 1).
Therefore, f6 (x, y, z) has the highest coefficient
We have
f6 (x, 1, 1) = 2x(x + 1)(x − 1)2 (x − k)2 ≥ 0,
4x 2 (x − 1)4 (x − k)2
Eα,β (x) = f k,0 (x) = .
9(4 − k)2 (x + 2)2
The condition (a) in Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Af k,0 (x) for x ∈ [0, 4].
We have
4(k2 − 20k + 40)x 2 (x − 1)4 (x − k)2
Af k,0 (x) = ,
(4 − k)2 (x + 2)2
2x(x − 1)2 (x − k)2 g(x)
f6 (x, 1, 1) − Af k,0 (x) = ,
(4 − k)2 (x + 2)2
where
g(x) = (4 − k)2 (x + 1)(x + 2)2 − 2(k2 − 20k + 40)x(x − 1)2 .
Since
(x + 1)(x + 2)2 − 5x(x − 1)2 = (4 − x)(1 + x + 4x 2 ) ≥ 0,
we get
g(x) ≥ 5(4 − k)2 x(x − 1)2 − 2(k2 − 20k + 40)x(x − 1)2 = 3k2 x(x − 1)2 ≥ 0.
( y − z)2 f ( y, z) ≥ 0,
where
f ( y, z) 1 1
= (2k − 3)2 − (2k − 3)2 + (αk − 3k − 1)
y 2z2 8 4
4k2 + 12k + 17
= αk − .
8
4k2 + 12k + 17
From the necessary condition f ( y, z) ≥ 0, we get αk ≥ .
8
A = −12(−27) = 324.
Since
f6 (x, 1, 1) = x 2 (x + 2)2 (x − 1)2 ,
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Since
Af0,−2 (x) = 4x 2 (x − 1)4 ,
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
that is
A1 = P1 (1, 1, 1) = −9(k2 + k + 1).
Therefore, f6 (x, y, z) has the highest coefficient
We have
f6 (x, 1, 1) = 2x(x − 1)2 (x − k)2 ≥ 0,
4x 2 (x − 1)4 (x − k)2
Eα,β (x) = f k,0 (x) = .
9(4 − k)2 (x + 2)2
The condition (a) of Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Af k,0 (x) for x ∈ [0, 4].
We have
4(8 − 4k − k2 )x 2 (x − 1)4 (x − k)2
Af k,0 (x) = ,
(4 − k)2 (x + 2)2
2x(x − 1)2 (x − k)2 g(x)
f6 (x, 1, 1) − Af k,0 (x) = ,
(4 − k)2 (x + 2)2
where
g(x) = (4 − k)2 (x + 2)2 − 2(8 − 4k − k2 )x(x − 1)2 .
Since
(x + 2)2 − x(x − 1)2 = (4 − x)(1 + x + x 2 ) ≥ 0,
Highest Coefficient Cancellation Method for Nonnegative Variables 395
we get
g(x) ≥ (4 − k)2 x(x − 1)2 − 2(8 − 4k − k2 )x(x − 1)2 = 3k2 x(x − 1)2 ≥ 0.
becomes
yz( y − z)[ y 3 − z 3 − k yz( y − z)] + αk y 2 z 2 ( y − z)2 ] ≥ 0,
yz( y − z)2 [ y 2 + z 2 + (1 − k + αk ) yz] ≥ 0.
The necessary condition
y 2 + z 2 + (1 − k + αk ) yz ≥ 0
leads to αk ≥ k − 3 for y = z = 1.
Solution. Denote
f (x, y, z) = x 2 (x − y)(x − z)(x − k y)(x − kz),
From
that is,
A1 = P1 (1, 1, 1) = 9(k2 + k + 1).
Since the highest coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is equal to −27,
f6 (x, y, z) has the highest coefficient
becomes as follows:
( y − z)2 f ( y, z) ≥ 0,
where
f ( y, z) = y 4 + z 4 − (k − 1) yz( y 2 + z 2 ) + (αk − k + 1) y 2 z 2 ≥ 0.
which is equivalent to
(x 2 + y 2 + z 2 − x y − yz − z x)(x 2 + y 2 + z 2 − 2x y − 2 yz − 2z x)2 ≥ 0.
Actually, this inequality is valid for all real x, y, z (see P 3.49 from chapter 3).
Observation 4. Substituting k − 1 for k in P 4.12, and using then the identity
x 2 (x − y)(x − z)[x − (k − 1) y](x − (k − 1)z] =
= x 2 (x − y)(x − z)(x − k y + z)(x − kz + y) + k(x − y)2 ( y − z)2 (z − x)2 ,
with equality for x = y = z, for x = 0 and y = z (or any cyclic permutation), and for
x/(k − 1) = y = z (or any cyclic permutation) if k = 1.
Solution. If one of x, y, z is zero, the inequality is trivial. On the other hand, the
inequality remains unchanged by replacing x, y, z and k with 1/x, 1/ y, 1/z and
1/k, respectively. Therefore, it suffices to consider that
Since
that is,
A = P1 (1, 1, 1) = 9(k2 + k + 1).
On the other hand,
f6 (x, 1, 1) = (x − 1)2 (x − k)2 ,
Highest Coefficient Cancellation Method for Nonnegative Variables 399
f6 (0, y, z) = k2 y 3 z 3 .
Since
In order to prove the conditions (b) and (c) of Theorem 2, consider two cases:
0 ≤ k ≤ 1 and k ≤ −1.
We have
δ2 y 4 z 4
ĝ0,δ ( y, z) = ,
( y + z)2
9(k2 + k + 1)δ2
f6 (0, y, z) − Aĝ0,δ ( y, z) ≥ f (k) y 3 z 3 , f (k) = k2 − .
4
Choosing
−2k
δ= ,
3(k + 1)
we have
k2 (k2 + k + 1)δ2 k3
f (k) = k2 − = ≥ 0,
(k + 1) 2 (k + 1)2
therefore the condition (c) of Theorem 2 is satisfied.
400 Vasile Cîrtoaje
(x − 1)4
Fγ,δ (x) = f∞,∞ (x) = .
9(x + 2)2
The condition (b) of Theorem 2 is satisfied if f6 (x, 1, 1) ≥ Af∞,∞ (x) for x ≥ 4.
We have
(k2 + k + 1)(x − 1)4 k2 (x − 1)4
Af∞,∞ (x) = ≤ ,
(x + 2) 2 (x + 2)2
therefore
(x − 1)2 [(x + 2)2 (x − k)2 − k2 (x − 1)2 ]
f6 (x, 1, 1) − Af∞,∞ (x) ≥ .
(x + 2)2
It suffices to show that
The equality occurs for x = y = z, for y = z = 0 (or any cyclic permutation), and
for x/k = y = z (or any cyclic permutation) if k = 0. If k = 0, then the equality
holds also for x = 0 and y = z.
Observation 1. The coefficient (zero) of the product (x − y)2 ( y − z)2 (z − x)2 is the
best possible.
Setting x = 0, the inequality
yz(x − y)(x − z)(x − k y)(x − kz) + αk (x − y)2 ( y − z)2 (z − x)2 ≥ 0
becomes as follows:
k2 y 3 z 3 + αk y 2 z 2 ( y − z)2 ≥ 0.
Setting y = 1 and z = 0 in the necessary condition
k2 yz + αk ( y − z)2 ≥ 0,
we get αk ≥ 0.
Observation 2. Substituting k − 1 for k in P 4.13, and using then the identity
yz(x − y)(x − z)[x − (k − 1) y](x − (k − 1)z] =
= yz(x − y)(x − z)(x − k y + z)(x − kz + y) + k(x − y)2 ( y − z)2 (z − x)2 ,
with equality for x = y = z, for y = z = 0 (or any cyclic permutation), and for
x/(k − 1) = y = z (or any cyclic permutation) if k = 1.
402 Vasile Cîrtoaje
We have
A = −3(−27) = 81.
Since
f6 (x, 1, 1) = (2x + 1)x 2 (x − 1)2 ,
we apply Corollary 1 for
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Condition (a). We need to show that f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈ [0, 4]. We
have
Af0,−2 (x) = x 2 (x − 1)4 ,
Ax 2 = 81x 2 ,
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
We have
A = −9(−27) = 243.
Since
f6 (x, 1, 1) = (2x + 1)(x − 1)2 (x + 2)2 ,
f6 (0, y, z) = yz 4 y 2 z 2 + ( y − z)( y 3 − z 3 ) + 2 yz( y − z)2 − 9 y 2 z 2 ( y − z)2
= yz 4 y 2 z 2 + ( y − z)4 − 8 yz( y − z)2
2
= yz ( y − z)2 − 2 yz = yz( y 2 + z 2 − 4 yz)2 ,
y 2 z 2 ( y 2 + z 2 − 4 yz)2
fˆ1,∞ ( y, z) = ,
81( y + z)2
we apply Theorem 2 for
(x − 1)4 (x − 2)2
Eα,β (x) = f2,−2 (x) = ,
324
4(x − 1)6
Fγ,δ (x) = f1,∞ (x) =
81(x + 2)2
Condition (a). We need to show that f6 (x, 1, 1) ≥ Af2,−2 (x) for x ∈ [0, 4]. We
have
3(x − 1)4 (x − 2)2
Af2,−2 (x) = ,
4
(x − 1)2 f (x)
f6 (x, 1, 1)−Af2,−2 (x) = , f (x) = 4(2x +1)(x +2)2 −3(x −1)2 (x −2)2 .
4
Since (x + 2)2 ≥ (x − 2)2 and 2x + 1 ≥ (x − 1)2 , we have f (x) > 0.
Condition (b). This condition is satisfied if f6 (x, 1, 1) ≥ Af1,∞ (x) for x ≥ 4. We
have
12(x − 1)6
Af1,∞ (x) = ,
(x + 2)2
(x − 1)2 f (x)
f6 (x, 1, 1) − Af1,∞ (x) = ,
(x + 2)2
where
f (x) = (2x + 1)(x + 2)4 − 12(x − 1)4 .
The necessary inequality f (x) ≥ 0 can be obtained by multiplying the inequalities
3 y 2 z 2 ( y 2 + z 2 − 4 yz)2
A fˆ1,∞ ( y, z) = ,
( y + z)2
3 yz
f6 (0, y, z) − A fˆ1,∞ ( y, z) = yz( y 2 + z 2 − 4 yz)2 1 − ≥ 0.
( y + z)2
The equality occurs for x = y = z, and for x = 0 and y/x + x/ y = 4 (or any
cyclic permutation).
We have
A = −6(−27) = 162.
Since
f6 (x, 1, 1) = (2x + 1)(x − 1)2 (x − 3)2 ,
f6 (0, y, z) = yz 9 y 2 z 2 + ( y − z)( y 3 − z 3 ) − 3 yz( y − z)2 − 6 y 2 z 2 ( y − z)2
2 2
= yz 9 y z + ( y − z)4 − 6 yz( y − z)2
2
= yz ( y − z)2 − 3 yz = yz( y 2 + z 2 − 5 yz)2 ,
y 2 z 2 ( y 2 + z 2 − 5 yz)2
fˆ3/2,∞ ( y, z) = ,
144( y + z)2
we apply Theorem 2 for
Condition (a). We need to show that f6 (x, 1, 1) ≥ Af3,−2 (x) for x ∈ [0, 4]. We
have
8(x − 1)4 (x − 3)2
Af3,−2 (x) = ,
25
(x − 1)2 (x − 3)2 f (x)
f6 (x, 1, 1) − Af3,−2 (x) = ,
25
with
f (x) = 25(2x + 1) − 8(x − 1)2 > 8[2x + 1 − (x − 1)2 ] ≥ 0.
Condition (b). This condition is satisfied if f6 (x, 1, 1) ≥ Af3/2,∞ (x) for x ≥ 4. We
have
9(x − 1)4 (2x − 3)2
Af3/2,∞ (x) = ,
8(x + 2)2
(x − 1)2 f (x)
f6 (x, 1, 1) − Af3/2,∞ (x) = ,
8(x + 2)2
where
f (x) = 8(2x + 1)(x − 3)2 (x + 2)2 − 9(x − 1)2 (2x − 3)2 .
Since 8(2x + 1) ≥ 72 > 64, we have f (x) ≥ 0 if
8(x − 3)(x + 2) ≥ 3(x − 1)(2x − 3).
Indeed,
8(x − 3)(x + 2) − 3(x − 1)(2x − 3) = 2x 2 + 7x − 57 ≥ 32 + 28 − 57 > 0.
The equality occurs for x = y = z, for x/3 = y = z (or any cyclic permutation),
and for x = 0 and y/x + x/ y = 5 (or any cyclic permutation).
We have
A = 27αk > 0,
f6 (x, 1, 1) = (2x + 1)(x − 1)2 (x − k)2 ,
f6 (0, y, z) = yz k2 y 2 z 2 + ( y − z)( y 3 − z 3 ) − k yz( y − z)2 − αk y 2 z 2 ( y − z)2
2
= yz k2 y 2 z 2 + ( y − z)4 − 2|k| yz( y − z)2 = yz ( y − z)2 − |k| yz
2
= yz ( y + z)2 − (4 + |k|) yz ,
y 2z2 2
ĝγ,δ ( y, z) = γ( y + z)2 + δ yz .
( y + z)2
We will apply Theorem 2 for
Conditions (b) and (c). To prove the conditions (b) and (c) of Theorem 2, we
consider four cases:
56
k ≤ 0, k ≥ 2, 0 ≤ k ≤ 1, 1≤k≤ .
25
In the first three cases, we choose
2
(2x + 1)2
Fγ,δ (x) = gγ,δ (x) = x + γ(x + 2)(2x + 1) + δ .
x +2
Highest Coefficient Cancellation Method for Nonnegative Variables 407
δ = −(4 + |k|)γ,
to have
γ2 y 2 z 2
ĝγ,δ ( y, z) = [( y + z)2 − (4 + |k|) yz]2 .
( y + z)2
In addition,
2
(4 + |k|)γ(2x + 1)2
gγ,δ (x) = x + γ(x + 2)(2x + 1) − .
x +2
Case 1: k ≤ 0. Choosing
2
γ=
,
27 1 − k
−2(4 − k)
δ = −(4 − k)γ =
,
27 1 − k
we have
A = 27αk = 81(1 − k),
f6 (x, 1, 1) − Agγ,δ (x) = (2x + 1)(x − 1)2 (x − k)2
2
2 2(4 − k) (2x + 1)2
−81(1 − k) x +
(x + 2)(2x + 1) −
· ,
27 1 − k 27 1 − k x +2
2(2x + 1)2
9(x − 1) ≥ .
x +2
We have
2(2x + 1)2 (x − 4)(x + 5)
9(x − 1) − = ≥ 0.
x +2 x +2
The condition (c) of Theorem 2 is satisfied if f6 (0, y, z) ≥ Aĝγ,δ ( y, z) for y, z ≥ 0.
We have
γ2 y 2 z 2
ĝγ,δ ( y, z) = [( y + z)2 − (4 + |k|) yz]2
( y + z)2
4 y 2z2
= [( y + z)2 − (4 − k) yz]2 ,
729(1 − k)( y + z)2
4 y 2z2 yz
Aĝγ,δ ( y, z) =
[( y+z)2 −(4−k) yz]2 ≤
[( y+z)2 −(4−k) yz]2 ,
9 1 − k( y + z)2 9 1−k
1
f6 (0, y, z) − Aĝγ,δ ( y, z) ≥ 1 −
yz[( y + z)2 − (4 − k) yz]2 ≥ 0.
9 1−k
Case 2: k ≥ 2. From the condition gγ,δ (k) = 0, where
2
(4 + k)γ(2x + 1)2
gγ,δ (x) = x + γ(x + 2)(2x + 1) − ,
x +2
Highest Coefficient Cancellation Method for Nonnegative Variables 409
we get
k+2
γ= ,
(2k + 1)(k + 5)
−(k + 2)(k + 4)
δ = −(4 + k)γ = ,
(2k + 1)(k + 5)
therefore
2
(k + 2)(x + 2)(2x + 1) (4 + k)(k + 2)(2x + 1)2
gγ,δ (x) = x + −
(2k + 1)(k + 5) (2k + 1)(k + 5)(x + 2)
g12 (x)
= ,
(2k + 1)2 (k + 5)2 (x + 2)2
where
g1 (x) = (2k + 1)(k + 5)x(x + 2) + (k + 2)(x + 2)2 (2x + 1) − (k + 2)(k + 4)(2x + 1)2
= (x − k)[2(k + 2)x 2 − (5k + 9)x + k + 4].
Since
A = 27αk = 27(k + 3),
the condition (b), namely f6 (x, 1, 1) ≥ Agγ,δ (x) for x ≥ 4, is true if
(2k +1)2 (k +5)2 (2x +1)(x −1)2 (x +2)2 ≥ 27(k +3)[2(k +2)x 2 −(5k +9)x + k +4]2 .
9(2k + 1)2 (k + 5)2 (x − 1)2 (x + 2)2 ≥ 27(k + 3)[2(k + 2)x 2 − (5k + 9)x + k + 4]2 .
In addition, because of
2(k +2)x 2 −(5k +9)x + k +4 ≥ 8(k +2)x −(5k +9)x + k +4 = (3k +7)x + k +4 > 0,
which is equivalent to
7(k + 7)x − 5k − 12 ≥ 0.
410 Vasile Cîrtoaje
γ2 y 2 z 2
ĝγ,δ ( y, z) = [( y + z)2 − (4 + k) yz]2 ,
( y + z)2
27(k + 3)γ2 y 2 z 2
Aĝγ,δ ( y, z) = [( y + z)2 − (4 + k) yz]2
( y + z)2
27(k + 3)γ2 yz
≤ [( y + z)2 − (4 + k) yz]2
4
27(k + 3)γ2
f6 (0, y, z) − Aĝγ,δ ( y, z) ≥ 1 − yz[( y + z)2 − (4 + k) yz]2 ,
4
we choose
2
γ=
3 3(k + 3)
to have f6 (0, y, z) − Aĝγ,δ ( y, z) ≥ 0 for y, z ≥ 0. Thus, the condition (c) of Theorem
2 is satisfied.
Highest Coefficient Cancellation Method for Nonnegative Variables 411
10(2x + 1)2
g(x) > 9x + 2(x + 2)(2x + 1) −
x +2
4x 3 − 13x 2 + 2x − 2
=
x +2
4x 2 (x − 4) + 3x 2 + 2x − 2
= > 0.
x +2
According to the AM-GM inequality
3 + (k + 3) ≥ 2 3(k + 3),
4x 2 − 5x + 10
(x − 1) 2x + 1 ≥ . (*)
x +2
412 Vasile Cîrtoaje
It is true if
(x − 1)2 (2x + 1)(x + 2)2 ≥ (4x 2 − 5x + 10)2 ,
which is equivalent to (x − 4)g3 (x) ≥ 0, where
which is true.
Case 4: 1 ≤ k ≤ 3. We choose
4(x − 1)4 (x − γ)2
Fγ,δ (x) = fγ,∞ (x) = .
9(1 + 2γ)2 (x + 2)2
k
γ=
2
to have
y 2 z 2 [( y + z)2 − (4 + k) yz]2
fˆk/2,∞ ( y, z) = ,
9(1 + k)2 ( y + z)2
3(k + 3) y 2 z 2 [( y + z)2 − (4 + k) yz]2
A fˆk/2,∞ ( y, z) = ,
(1 + k)2 ( y + z)2
therefore,
2
3(k + 3) yz
f6 (0, y, z) − A fˆk/2,∞ ( y, z) = 1 − yz ( y + z)2 − (4 + k) yz ≥ 0.
(1 + k)2 ( y + z)2
Thus, the condition (c) of Theorem 2 is satisfied. With regard to the condition (b),
we have
(x − 1)4 (2x − k)2
f k/2,∞ (x) = ,
9(1 + k)2 (x + 2)2
3(k + 3)(x − 1)4 (2x − k)2
Af k/2,∞ (x) = ,
(1 + k)2 (x + 2)2
(x − 1)2 f (x)
f6 (x, 1, 1) − Af k/2,∞ (x) = ,
(1 + k)2 (x + 2)2
f (x) = (1 + k)2 (2x + 1)(x + 2)2 (x − k)2 − 3(k + 3)(x − 1)2 (2x − k)2 .
Highest Coefficient Cancellation Method for Nonnegative Variables 413
Since
(2x + 1)(x + 2)2 − 12(x + 1)2 ≥ 6(2x + 1)(x + 2) − 12(x + 1)2 = 6x > 0,
we get
f (x) > 12(1 + k)2 (x + 1)2 (x − k)2 − 3(k + 3)(x − 1)2 (2x − k)2 .
In addition, since
4 + (k + 3) ≥ 4 k + 3,
it suffices to show that
Since
6(k − 1)x 2 ≥ 24(k − 1)x,
it suffices to show that
which is
(−2 + 33k − 7k2 )x − 7 − 9k − 8k2 ≥ 0.
Since
−2 + 33k − 7k2 > −9 + 30k − 9k2 = 3(3 − k)(3k − 1) ≥ 0,
we have
The equality holds for x = y = z, for x = 0 and y/z + z/ y = 2 + |k| (or any cyclic
permutation), and for x/k = y = z (or any cyclic permutation) if k > 0
Observation. The coefficient αk of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, the inequality turns into
k(αk − k − 3) y 2 z 2 ≤ 0,
which involves αk ≤ k + 3.
For k < 0, choosing y 2 + z 2 = (2 − k) yz, we get
(−k)(αk + 3k − 3) y 2 z 2 ≤ 0,
which yields α0 ≤ 3.
We have
A = −4(−27) = 108.
Since
f6 (x, 1, 1) = 2x 2 (x + 2)(x − 1)2 ,
we apply Corollary 1 for
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Since
2x 2 (x − 1)2 (2x + 1)(4 − x)
f6 (x, 1, 1) − Af0,−2 (x) = ≥0
3
for 0 ≤ x ≤ 4, the condition (a) in Corollary 1 is satisfied.
The condition (b) in Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Ax 2 for x ≥ 4. This is
true since
f6 (x, 1, 1) − Ax 2 = x 2 2(x + 2)(x − 1)2 − 108 ≥ x 2 (2 · 6 · 9 − 108) = 0.
Highest Coefficient Cancellation Method for Nonnegative Variables 415
The equality occurs for x = y = z, for x = 0 and y = z (or any cyclic permutation),
and for y = z = 0 (or any cyclic permutation).
We have
A = 27(2 + |k|)2 > 0,
f6 (x, 1, 1) = 2(x + 2)(x − 1)2 (x − k)2 ,
f6 (0, y, z) = ( y + z) k2 y 2 z 2 ( y + z) + yz( y − z)( y 2 − z 2 ) − (2 + |k|)2 y 2 z 2 ( y − z)2
= k2 y 2 z 2 ( y + z)2 + yz( y − z)2 ( y + z)2 − (2 + |k|)2 yz
= k2 y 2 z 2 ( y + z)2 + yz ( y + z)2 − 4 yz ( y + z)2 − (2 + |k|)2 yz
= yz( y + z)4 − 4(2 + |k|) y 2 z 2 ( y + z)2 + 4(2 + |k|)2 y 3 z 3
2
= yz ( y + z)2 − 2(2 + |k|) yz ,
y 2z2 2
ĝγ,δ ( y, z) = γ( y + z)2 + δ yz .
( y + z)2
We will apply Theorem 2 for
Conditions (b) and (c). To prove the conditions (b) and (c) of Theorem 2, we
consider two cases:
|k| ≥ 1, |k| ≤ 1.
γ = |k|
to have
y 2 z 2 [( y + z)2 − 2(2 + |k|) yz]2
fˆ|k|,∞ ( y, z) = ,
9(1 + 2|k|)2 ( y + z)2
3(2 + |k|)2 y 2 z 2 [( y + z)2 − 2(2 + |k|) yz]2
A fˆ|k|,∞ ( y, z) = ,
(1 + 2|k|)2 ( y + z)2
therefore,
3(2 + |k|)2 yz 2
f6 (0, y, z) − A fˆ|k|,∞ ( y, z) = 1 − yz ( y + z)2 − (2 + |k|)2 yz .
(1 + 2|k|)2 ( y + z)2
Indeed,
which is equivalent to
This is true for x ≥ 4 and |k| ≥ 1 because 1 + 2|k| ≥ 2 + |k| and x + 2 > x − 1.
δ = −2(2 + |k|)γ,
to have
γ2 y 2 z 2
ĝγ,δ ( y, z) = [( y + z)2 − 2(2 + |k|) yz]2 .
( y + z)2
In addition, by choosing
1
γ= ,
3(2 + |k|)
we have
y 2z2
ĝγ,δ ( y, z) = [( y + z)2 − 2(2 + |k|) yz]2 ,
9(2 + |k|)2 ( y + z)2
3 y 2z2
Aĝγ,δ ( y, z) = [( y + z)2 − 2(2 + |k|) yz]2 ,
( y + z)2
2
3 yz
f6 (0, y, z) − Aĝγ,δ ( y, z) = 1 − yz ( y + z)2 − 2(2 + |k|)2 yz ≥ 0.
( y + z)2
Thus, the condition (c) of Theorem 2 is satisfied.
418 Vasile Cîrtoaje
f6 (x, 1, 1) − Agγ,δ (x) ≥ 12(x − 1)2 (x − |k|)2 − Agγ,δ (x) = 3g1 (x)g2 (x),
where
(2 + |k|)(5x 2 + 2x + 2)
g1 (x) = 2(x − 1)(x − |k|) − (x + 2)(2x + 1) +
x +2
3[(1 + |k|)x 2 − 4x + 2|k|] 3(x 2 − 4x)
= ≥ ≥ 0.
x +2 x +2
(2 + |k|)(5x 2 + 2x + 2)
g2 (x) = 2(x − 1)(x − k) + (x + 2)(2x + 1) −
x +2
3(5x 2 + 2x + 2)
≥ 2(x − 1)(x − 1) + (x + 2)(2x + 1) −
x +2
5x 2 + 2x + 2
≥ 2(x − 1) + (x + 2)(2x + 1) −
2
2
5x 2 + 2x + 2 3(x 2 + 2)
= 4x + x + 4 −
2
= > 0.
2 2
The equality occurs for x = y = z, for x/k = y = z (or any cyclic permutation)
if k = 0, for y = z = 0 (or any cyclic permutation), and for x = 0 and y/z + z/ y =
2 + 2|k| (or any cyclic permutation).
Observation. The coefficient of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, the inequality
αk (x − y)2 ( y − z)2 (z − x)2
( y + z)(x − y)(x − z)(x − k y)(x − kz) ≥
x + y +z
becomes
yz ( y 2 + z 2 )2 − (αk − k2 ) yz( y 2 + z 2 ) + 2(αk + k2 − 2) y 2 z 2 ≥ 0.
|k|[αk − (2 + |k|)2 ] y 2 z 2 ≤ 0,
A = −12(−27) = 324.
Since
f6 (x, 1, 1) = x(x + 2)(x 2 − 1)2 ,
we apply Corollary 1 for
4x 2 (x − 1)4 (x + 1)2
Eα,β (x) = f0,−1 (x) = .
225(x + 2)2
Since
x(x − 1)2 (x + 1)2 f (x)
f6 (x, 1, 1) − Af0,−1 (x) = ,
225(x + 2)2
where
f (x) = 225(x + 2)3 − 1296x(x − 1)2 > 216[(x + 2)3 − 6x(x − 1)2 ],
Indeed, we have
4
Then,
αk (x − y)2 ( y − z)2 (z − x)2
x(x − y)(x − z)(x − k y)(x − kz) + ≥ 0.
x + y +z
(Vasile C., 2010)
Solution. Write the inequality as f6 (x, y, z) ≥ 0, where
f6 (x, y, z) = (x + y+z) x(x − y)(x −z)(x −k y)(x −kz)+αk (x − y)2 ( y−z)2 (z−x)2 .
Since the product (x − y)2 ( y − z)2 (z − x)2 has the highest coefficient equal to −27,
f6 (x, y, z) has the highest coefficient
A = −27αk .
Also, we have
f6 (x, 1, 1) = x(x + 2)(x − 1)2 (x − k)2 ,
f6 (0, y, z) = ( y − z)2 [( y + z)4 − (k + 2) yz( y + z)2 + αk y 2 z 2 ].
There are three cases to consider.
Case 1: k ≥ 6. Since
αk = (k + 2)2 ,
−27(k + 2)2
A = −27αk = < 0,
4
the desired inequality is true if f6 (x, 1, 1) ≥ 0 and f6 (0, y, z) ≥ 0 for x, y, z ≥ 0
(Theorem 1). The first condition is clearly true and
(k + 2)2 2 2
f6 (0, y, z) = ( y − z)2 [( y + z)4 − (k + 2) yz( y + z)2 + y z ]
4
2
k+2
= ( y − z)2 ( y + z)2 − yz ≥ 0.
2
Highest Coefficient Cancellation Method for Nonnegative Variables 421
Case 2: 2 ≤ k ≤ 6. Since
αk = 4(k − 2),
A = −27αk = −108(k − 2) ≤ 0,
the desired inequality is true if f6 (x, 1, 1) ≥ 0 and f6 (0, y, z) ≥ 0 for x, y, z ≥ 0
(Theorem 1). The first condition is true and
Case 3: k ≤ 2. We have
αk = 4(k − 2),
A = −27αk = 108(2 − k) ≥ 0.
We will apply Corollary 1 for
4x 2 (x − 1)4 (x − k)2
Eα,β (x) = f k,0 (x) = .
9(4 − k)2 (x + 2)2
The condition (a) in Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Af k,0 (x) for x ∈ [0, 4].
We have
48(2 − k)x 2 (x − 1)4 (x − k)2
Af k,0 (x) = ,
(4 − k)2 (x + 2)2
x(x − 1)2 (x − k)2 [(4 − k)2 (x + 2)3 − 48(2 − k)x(x − 1)2 ]
f6 (x, 1, 1) − Af k,0 (x) = .
(4 − k)2 (x + 2)2
The condition (a) is true if
(4 − k)2 ≥ 8(2 − k)
and
(x + 2)3 ≥ 6x(x − 1)2 ,
which are equivalent to k2 ≥ 0 and (4 − x)(2 + 2x + 5x 2 ) ≥ 0, respectively.
The condition (b) in Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Ax 2 for x ≥ 4. It
suffices to show that
Indeed, we have
4(x − 1)2 − 9x = (x − 4)(4x − 1) ≥ 0,
The equality occurs for x = y = z, for x = 0 and y = z (or any cyclic permutation),
and for x/k = y = z (or any cyclic permutation) if k = 0, and for x = 0 and
y/z + z/ y = (k − 2)/2 (or any cyclic permutation) if k > 6
Observation. The coefficient αk of the product (x − y)2 ( y − z)2 (z − x)2 is the best
possible. Setting x = 0, the inequality becomes
becomes
[4αk − (k + 2)2 ] y 2 z 2 ≥ 0,
We have
A = −5(−27) = 135,
f6 (x, 1, 1) = (2x + 1)(x 2 + 1)(x − 1)2 ,
f6 (0, y, z)
= y 2 z 2 + ( y − z)2 ( y 2 + z 2 + yz) − 5 yz( y − z)2
yz
= y 2 z 2 + ( y 2 + z 2 − 2 yz) y 2 + z 2 − 4 yz
= ( y 2 + z 2 )2 − 6( y 2 + z 2 ) yz + 9 y 2 z 2
= ( y 2 + z 2 − 3 yz)2 ,
y 2 z 2 ( y 2 + z 2 − 3 yz)2
fˆ1/2,∞ ( y, z) = .
36( y + z)2
The conditions (a) and (b) in Theorem 2 are satisfied if f6 (x, 1, 1) ≥ Af1/2,∞ (x)
for x ≥ 0. Since
15(x − 1)4 (2x − 1)
Af1/2,∞ (x) = ,
4(x + 2)2
(x − 1)2 f (x)
f6 (x, 1, 1) − Af1/2,∞ (x) = ,
4(x + 2)2
where
f (x) = 4(x + 2)2 (2x + 1)(x 2 + 1) − 15(x − 1)2 (2x − 1)2 ,
we need to show that f (x) ≥ 0 for x ≥ 0. This is true if
that is
x(2x 2 − 7x + 28) ≥ 0.
15 y 2 z 2 ( y 2 + z 2 − 3 yz)2
A fˆ1/2,∞ ( y, z) = ,
4( y + z)2
424 Vasile Cîrtoaje
15 yz
f6 (0, y, z) − A fˆ1/2,∞ ( y, z) = yz( y 2 + z 2 − 3 yz)2 1 −
4( y + z)2
15
≥ yz( y 2 + z 2 − 3 yz)2 1 − ≥ 0.
16
The equality occurs for x = y = z, and for x = 0 and y/z + z/ y = 3 (or any
cyclic permutation).
We have
A = −16(−27) = 432,
f6 (x, 1, 1) = (2x + 1)(4x 2 + 1)(x − 1)2 ,
f6 (0, y, z)
= y 2 z 2 + 4( y − z)2 ( y 2 + z 2 + yz) − 16 yz( y − z)2
yz
= y 2 z 2 + 4( y 2 + z 2 − 2 yz) y 2 + z 2 − 3 yz
= 4( y 2 + z 2 )2 − 20( y 2 + z 2 ) yz + 25 y 2 z 2
= (2 y 2 + 2z 2 − 5 yz)2 .
Apply Theorem 2 for
3 3
to have g( y, z) ≥ 0. Thus, the condition (c) in Theorem 2 is satisfied.
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Agγ,δ (x) for x ≥ 4, where
2
9γ(2x + 1)2
gγ,δ (x) = x + γ(x + 2)(2x + 1) − .
2(x + 2)
Since
2
27 3γ(2x + 1)2
Agγ,δ (x) = 4 6 3x + 6 3γ(x + 2)(2x + 1) − = 4 f 2 (x),
x +2
9(2x + 1)2
f (x) = 6 3x + 2(x + 2)(2x + 1) − ,
x +2
we need to show that
for x ≥ 4. Since
9(2x + 1)2
f (x) > 9x + 2(x + 2)(2x + 1) −
x +2
x 2 (4x − 9) + 6x − 1
= > 0,
x +2
it suffices to show that
where
9(2x + 1)2
f1 (x) = 11x + 2(x + 2)(2x + 1) − > f (x).
x +2
Since
(2x + 1)(4x 2 + 1)(x − 1)2 > (2x + 1)(4x 2 )(x − 1)2
and
4x 3 − 7x 2 + 10x − 1 x(4x 2 − 7x + 10)
f1 (x) = < ,
x +2 x +2
it suffices to show that
Since
(x + 5)2
2x + 1 ≤ ,
9
it suffices to show that
which is equivalent to
x 3 − 2x 2 + 8x − 28 ≥ 0,
(x − 4)(x 2 + 2x + 16) + 36 ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y/z + z/ y = 5/2 (or any
cyclic permutation).
Highest Coefficient Cancellation Method for Nonnegative Variables 427
We have
A = 27(3 + 2 k),
f6 (x, 1, 1) = (2x + 1)(x 2 + k)(x − 1)2 ,
f6 (0, y, z)
= k y 2 z 2 + ( y − z)2 ( y 2 + z 2 + yz) − (3 + 2 k) yz( y − z)2
yz
" #
= k y 2 z 2 + ( y 2 + z 2 − 2 yz) y 2 + z 2 − 2(1 + k) yz
= ( y 2 + z 2 )2 − 2(2 + k)( y 2 + z 2 ) yz + (2 + k)2 y 2 z 2
" #2
= y 2 + z 2 − (2 + k) yz ,
f6 (x, 1, 1) − Af
k,−2 (x) = (x − 1)2 f (x),
where
4 3+2 k 2
f (x) = (2x + 1)(x 2 + k) −
2 (x − 1) x − k ,
2
3 2+ k
the condition (a) of Theorem 2 is satisfied if f (x) ≥ 0 for 0 ≤ x ≤ 4. This is true
because
4 3+2 k
1≥
2 ,
3 2+ k
2x + 1 ≥ (x − 1)2 ,
428 Vasile Cîrtoaje
2
x2 + k ≥ x − k .
Indeed,
2x + 1 − (x − 1)2 = x(4 − x) ≥ 0,
2
x 2 + k − x − k = 2 k x ≥ 0.
With regard to the conditions (b) and (c), we consider two cases: 0 ≤ k ≤ 1 and
k ≥ 1.
Case 1: k ≥ 1. Having in view the expression of f6 (0, y, z), we will apply Theorem
2 for
2
(x − 1)4 2x − k
Fγ,δ (x) = f k/2,∞ (x) =
2 ,
9 1 + k (x + 2)2
2
y 2 z 2 y 2 + z 2 − (2 + k) yz
ˆ
f k/2,∞ ( y, z) =
.
9(1 + k )2 ( y + z)2
The condition (c) is satisfied because
" #2
f6 (0, y, z) − A fˆ
k/2,∞ ( y, z) = yz y 2 + z 2 − (2 + k) yz f ( y, z),
where
3(3 + 2 k ) yz
f ( y, z) = 1 −
(1 + k )2 ( y + z)2
3(3 + 2 k ) 4k + 2 k − 5
≥1−
=
> 0.
4(1 + k )2 4(1 + k )2
Since
3(3 + 2 k )
2 < 4
1+ k
and 2
2x − k ≤ 4x 2 + k,
Highest Coefficient Cancellation Method for Nonnegative Variables 429
x(2x 2 − 7x + 12) ≥ 0.
Since
" #2
f6 (0, y, z) − Aĝγ,δ ( y, z) = yz y 2 + z 2 − (2 + k ) yz g( y, z),
where
3 3(3 + 2 k )
to have g( y, z) ≥ 0. Thus, the condition (c) in Theorem 2 is satisfied.
The condition (b) is satisfied if f6 (x, 1, 1) ≥ Agγ,δ (x) for x ≥ 4, where
2
(2x + 1)2
gγ,δ (x) = x + γ(x + 2)(2x + 1) − (4 + k )γ .
x +2
430 Vasile Cîrtoaje
Since
4
Agγ,δ (x) =
gγ,δ (x) = f 2 (x),
γ2
2x (2x + 1)2
f (x) = + 2(x + 2)(2x + 1) − 2(4 + k) ,
γ x +2
we need to show that
10(2x + 1)2
f (x) ≥ 9x + 2(x + 2)(2x + 1) −
x +2
x 2 (4x − 13) + 2(x − 1)
= > 0.
x +2
On the other hand, by the AM-GM inequality,
-
2 3
= 3 3(3 + 2 k) ≤ [3 + (3 + 2 k)] = 3(3 + k).
γ 2
Therefore,
(2x + 1)2
f (x) ≤ 3(3 + k)x + 2(x + 2)(2x + 1) − 2(4 + k)
x +2
x(4x 2 − 5x + 10) − k(5x 2 + 2x + 2)
=
x +2
x(4x 2 − 5x + 10)
≤ .
x +2
In addition,
(2x + 1)(x 2 + k)(x − 1)2 ≥ (2x + 1)x 2 (x − 1)2 .
Thus, it suffices to prove that
According to
(x + 5)2
2x + 1 ≤ ,
9
it suffices to show that
1
(2x + 1)(x − 1)(x + 2) ≥ (x + 5)(4x 2 − 5x + 10).
3
Highest Coefficient Cancellation Method for Nonnegative Variables 431
(x − 4)(x 2 + x + 7) ≥ 0.
In addition, for
y 2 + z 2 = (2 + k) yz,
we get the necessary condition
3 + 2 k − αk y 2 z 2 ( y − z)2 ≥ 0,
which involves αk ≤ 3 + 2 k.
C = 4( 2 + 1)
and write the inequality as f6 (x, y, z) ≥ 0, where
f6 (x, y, z) = (x 2 − yz)2 (x − y)(x − z) − C(x − y)2 ( y − z)2 (z − x)2 .
!
Since (x − y)(x − z) = x 2 + 2 yz − q, the sum (x 2 − yz)2 (x − y)(x − z) has the
same highest coefficient A1 as
P1 (x, y, z) = (x 2 − yz)2 (x 2 + 2 yz)2 ,
We have
f6 (x, 1, 1) = (x + 1)2 (x − 1)4 ,
= ( y 2 + z 2 )3 − ( y 2 + z 2 )2 yz − (4 2 + 7)( y 2 + z 2 ) y 2 z 2 + (8 2 + 11) y 3 z 3
2
= y 2 + z 2 + (2 2 + 1) yz y 2 + z 2 − ( 2 + 1) yz .
In addition, for
2−1
γ= ,
2
we have
y 2 z 2 [ y 2 + z 2 − ( 2 + 1) yz]2
fˆγ,∞ ( y, z) = .
18( y + z)2
5(3x − 2)2
f (x) > 12(x + 1)2 −
2
5(3x − 2)2
> 10(x + 1) −
2
2
25x(4 − x)
= ≥ 0.
2
Highest Coefficient Cancellation Method for Nonnegative Variables 433
2−1
Condition (b). We need to show that f6 (x, 1, 1) ≥ Afγ,∞ (x) for γ = and
2
x ≥ 4. Since
6( 2 + 1) < 15,
it suffices to show that
6( 2 + 1) y 2 z 2 [ y 2 + z 2 − ( 2 + 1) yz]2
A fˆγ,∞ ( y, z) = ,
( y + z)2
2
2
f6 (0, y, z) − A fˆγ,∞ ( y, z) = y + z 2 − ( 2 + 1) yz f0 ( y, z),
where
6( 2 + 1) y 2 z 2
f0 ( y, z) = y 2 + z 2 + (2 2 + 1) yz −
( y + z)2
3( 2 + 1) yz ( 2 + 3) yz
≥ 2 yz + (2 2 + 1) yz − = ≥ 0.
2 2
4x 2 + y 2 + z 2 = 3x 2 + p2 − 2q,
we have
A = −9(3)3 = −243.
By Theorem 1, we only need to prove the original inequality for y = z = 1 and for
x = 0.
Case 1: y = z = 1. We need to show that
1 2 9
+ ≥ ,
4x 2 + 2 x 2 + 5 4x 2 + 4x + 10
which is equivalent to
x(x − 1)2 ≥ 0.
1 5( y 2 + z 2 ) 9
+ ≥ ,
y2 +z 2 4( y + z 4 ) + 17 y 2 z 2
4 4( y 2 + z 2 ) + 2 yz
For yz = 0, the inequality is an equality. For yz = 0, using the substitution
y 2 + z2
t= , t ≥ 2,
yz
yz 5 yz( y 2 + z 2 9 yz
+ ≥ ,
y2 +z 2 4( y + z 2 )2 + 9 y 2 z 2
2 4( y 2 + z 2 ) + 2 yz
Highest Coefficient Cancellation Method for Nonnegative Variables 435
1 5t 9
+ ≥ ,
t 4t 2 + 9 4t + 2
(t − 2)(2t − 1) ≥ 0,
The equality holds for x = y = z, for x = 0 and y = z (or any cyclic permutation),
and for y = z = 0 (or any cyclic permutation).
We have
f (t) ≥ f (2) ≥ 0
because
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation). If
k = 4, then the equality holds also for y = z = 0 (or any cyclic permutation).
For
(
f6 (x, y, z) = P (k y 2 + z 2 + x 2 )(kz 2 + x 2 + y 2 ) − 9(k + 5) (k x 2 + y 2 + z 2 ),
and
(k − 1)2 (t − 2)[2(4 − k)t 2 + 18t − (k − 1)2 ] ≥ 0,
where t ≥ 2.
P 4.27. If x, y, z are nonnegative real numbers, no two of which are zero, then
2 2 2 45
+ + ≥ .
x 2 + y 2 y 2 + z2 z2 + x 2 4(x 2 + y 2 + z 2 ) + x y + yz + z x
P = 4(x 2 + y 2 + z 2 ) + x y + yz + z x.
The highest coefficient
+ A of f6 (x, y, z) is equal with the highest coefficient of the
product −45 ( y 2 + z 2 ). Since
y 2 + z 2 = −x 2 + p2 − 2q,
we have
A = −45(−1)3 = 45.
In addition,
and
f6 (0, y, z) ≥ 0.
The last inequality is true if the original inequality holds for x = 0. Thus, we need
to show that
2 2 2 45
+ + ≥ ,
y 2 z2 y 2 + z2 4( y 2 + z 2 ) + yz
which can be written as
2( y 2 + z 2 ) 2 45
+ 2 ≥ ,
y 2z2 y + z2 4( y 2 + z 2 ) + yz
2( y 2 + z 2 ) 2 yz 45 yz
+ 2 ≥ .
yz y + z2 4( y 2 + z 2 ) + yz
Using the notation
y 2 + z2
t= , t ≥ 2,
yz
the inequality becomes
2 45
2t + ≥ ,
t 4t + 1
(t − 2)(8t + 18t − 1) ≥ 0.
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Since
x(x − 1)2 f (x)
f6 (x, 1, 1) − Af0,−2 (x) = ,
9
where
Second Solution. Apply Theorem 3. Since the conditions (a) and (c) are satisfied,
we only need to show that
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
f6 (x, 1, 1)
= P(x, 1, 1)[x 2 + k + 1 + 2(k x 2 + 2)] − 9(k + 5)(x 2 + k + 1)(k x 2 + 2)
x2 + k + 1
= 2(1 − k)2 x(x − 1)2 [(4 − k)x + 2k + 10] ≥ 0,
and
f6 (0, y, z) ≥ 0.
Highest Coefficient Cancellation Method for Nonnegative Variables 439
Since the conditions (a) and (c) in Theorem 3 are satisfied, it suffices to show that
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
4Ax(x − 1)3 2(1 − k)2 x(x − 1)2 f (x)
f6 (x, 1, 1) − = ,
27 3
where
P 4.28. If x, y, z are nonnegative real numbers, no two of which are zero, then
1 18
≥ .
2 y 2 + yz + 2z 2 5(x 2 + y 2 + z 2 + x y + yz + z x)
P(x, y, z) = x 2 + y 2 + z 2 + x y + yz + z x.
Since
2 y 2 + yz + 2z 2 = −2x 2 + yz + 2(p2 − 2q),
f6 (x, y, z) has the same highest coefficient A as
(
P3 (x, y, z) = −18 (−2x 2 + yz),
that is
A = P3 (1, 1, 1) = −18(−2 + 1)3 = 18.
We have
P(x, 1, 1) = x 2 + 2x + 3,
440 Vasile Cîrtoaje
f6 (x, 1, 1)
= 5(x 2 + 2x + 3)[(2x 2 + x + 2) + 10] − 90(2x 2 + x + 2)
2x 2 + x + 2
= 5x(2x 3 + 5x 2 − 16x + 9) = 5x(x − 1)2 (2x + 9),
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
2x 2 + x + 2 ≥ 2(x − 1)2 ,
45(2x 2 + x + 2) > x.
Second Solution. Apply Theorem 3. Since the conditions (a) and (c) are satisfied,
we only need to show that the condition (b) is satisfied. Thus, we need to prove
that
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
For
(
f6 (x, y, z) = P(x, y, z) (x 2 +k x y+ y 2 )(x 2 +k xz+z 2 )−9(2k+5) ( y 2 +k yz+z 2 ),
P(x, y, z)
= 2(2 − k)(x 2 + y 2 + z 2 ) + (4k + 1)(x y + yz + z x),
k+2
we have
A = −9(2k + 5)(k − 1)3 ,
P(x, 1, 1)
= 2(2 − k)x 2 + 2(4k + 1)x + 9,
k+2
f6 (x, 1, 1) = 2(k + 2)(x 2 + k x + 1)x(x − 1)2 [(2 − k)x + (k + 1)(5 − k)].
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
442 Vasile Cîrtoaje
x 2 + k x + 1 ≥ x 2 − x + 1 > (x − 1)2 ,
3(k + 2) ≥ 2k + 5,
4 ≥ (1 − k)2
and
3(2 − k) > 2(1 − k).
Condition (b). Since
f6 (x, 1, 1) − Ax 2 ≥ x 2 g(x),
g(x) = 2(k + 2)(2 − k)(x 2 + k x + 1)(x − 1)2 − 9(2k + 5)(1 − k)3 ,
we need to show that g(x) ≥ 0 for x ≥ 4. It suffices to show that g(4) ≥ 0, which
is true.
Condition (c). For x = 0, the original inequality becomes
k+2 1 1 9(2k + 5)
+ (k + 2) + ≥ ,
y 2 + k yz + z 2 y 2 z2 2(2 − k)( y 2 + z 2 ) + (4k + 1) yz
k+2 9(2k + 5)
+ (k + 2)t ≥ ,
t+k 2(2 − k)t + 4k + 1
(t − 2)h(t) ≥ 0,
where
h(t) = 2(4 − k2 )t 2 + (k + 2)(9 + 4k − 2k2 )t + 7k2 + 18k − 1,
h(t) ≥ h(2) = 67 + 52k − k2 − 4k3 > 0.
P 4.29. If x, y, z are nonnegative real numbers, no two of which are zero, then
x − (2 +
2)( y + z)
9(3 + 2 2)
+ ≥ 0.
( y + z)2 4(x + y + z)
A1 = 1.
We have
f (x, 1, 1)
= (x − 4 − 2 2)(x + 1)2 − 8 (2 + 2)x + 1 + 2
(x + 1)2
g(x) =(x + 2) x 3 − 2(1 + 2)x 2 − (23 + 12 2)x − 12 − 10 2
=x 4 − 2 2x 3 + 2 2x 2 − 2(2 − 2)x + 3 − 2 2
therefore
f (0, y, z)
= −(2 + 2) y 2 z 2 + ( y 2 + z 2 + 2 yz) y 2 + z 2 − (3 + 2) yz
y +z
= ( y 2 + z 2 )2 − (1 + 2)( y 2 + z 2 ) yz − (8 + 3 2) y 2 z 2 ,
f6 (0, y, z)
=4 ( y 2 + z 2 )2 − (1 + 2)( y 2 + z 2 ) yz − (8 + 3 2) y 2 z 2
( y + z) 2
+ 9(3 + 2 2) y 2 z 2
and
2
( y + z)2 2( y 2 + z 2 )2 − (3 + 2) yz
fˆ
2−1,−2 ( y, z) =
,
81(1 + 2)2
2
( y + z)2 2( y 2 + z 2 )2 − (3 + 2) yz
A fˆ
2−1,−2 ( y, z) = .
9
Therefore,
8( y + z)2 g( y, z)
f6 (0, y, z) − A fˆ
2−1,−2 ( y, z) = ,
9
where
g( y, z) = 4( y 2 + z 2 )2 − (3 + 4 2) yz( y 2 + z 2 ) + (6 2 − 7) y 2 z 2
= ( y 2 + z 2 − 2 yz) 4( y 2 + z 2 ) − (4 2 − 5) yz + (3 − 2 2) y 2 z 2 ≥ 0.
x
The equality occurs for x = y = z, and for
= y = z (or any cyclic
2−1
permutation).
Highest Coefficient Cancellation Method for Nonnegative Variables 445
P 4.30. If x, y, z are nonnegative real numbers, no two of which are zero, then
6x − y − z 6 y − z − x 6z − x − y 18
+ + ≥ .
y 2 + z2 z2 + x 2 x2 + y2 x + y +z
A1 = −1.
A = −18A1 = 18.
We have
f (x, 1, 1)
= 2(3x − 1)(x 2 + 1) + 4(5 − x) = 2(3x 3 − x 2 + x + 9),
x2 + 1
Also, we have
f6 (0, y, z) ≥ 0
for y, z ≥ 0. This is true if the original inequality holds for x = 0. Thus, we need
to show that
− y − z 6 y − z 6z − y 18
+ + ≥ ,
y 2 + z2 z2 y2 y +z
which is equivalent to
− y − z 6( y 3 + z 3 ) − yz( y + z) 18
+ ≥ ,
y 2 + z2 y 2z2 y +z
−1 6( y 2 + z 2 ) − 7 yz 18
+ ≥ ,
y2 +z 2 y 2z2 ( y + z)2
446 Vasile Cîrtoaje
− yz 6( y 2 + z 2 ) − 7 yz 18 yz
+ ≥ .
y +z
2 2 yz ( y + z)2
Using the substitution
y 2 + z2
t= , t ≥ 2,
yz
the inequality can be written as follows:
−1 18
+ 6t − 7 ≥ ,
t t +2
6t 3 + 5t 2 − 33t − 2 ≥ 0,
(t − 2)(6t 2 + 17t + 1) ≥ 0.
First Solution. We apply Theorem 3. The conditions (a) and (c) are clearly satis-
fied. The condition (b) is satisfied if
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
4Ax(x − 1)3 2x(x − 1)2 f (x)
f6 (x, 1, 1) − = ,
27 3
where
f (x) = 3(x 2 + 1)(3x + 11) − 4(x − 1) ≥ 6(3x + 11) − 4(x − 1) = 14(x + 5) > 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Second Solution. We apply Corollary 1 for
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
2(x − 1)4 x 2
Af0,−2 (x) = ,
9
2x(x − 1)2 g(x)
f6 (x, 1, 1) − Af0,−2 (x) = ≥ 0,
9
where
g(x) = 9(x 2 + 1)(3x + 11) − x(x − 1)2 .
Since x 2 + 1 ≥ (x − 1)2 , it suffices to show that
9(3x + 11) ≥ x,
Highest Coefficient Cancellation Method for Nonnegative Variables 447
where
f (x, y, z) = [(6 − 4k)x + (4k − 1)( y + z)](x 2 + k x y + y 2 )(x 2 + k xz + z 2 ),
we get
A = 36(k + 1)(1 − k)3 .
We have
f (x, 1, 1)
= [(3 − 2k)x + 4k − 1](x 2 + k x + 1) + (k + 2)[(4k − 1)x + 5]
2(x 2 + k x + 1)
= (3 − 2k)x 3 − (2k2 − 7k + 1)x 2 + (8k2 + 4k + 1)x + 9k + 9,
(t − 2)(B1 t 2 + B2 t + B3 ) ≥ 0,
where
Since B1 ≥ 0 and
we have
Case 1: k ∈ (−2, −1] ∪ [1, 3/2]. Since A ≤ 0, f6 (x, 1, 1) ≥ 0 and f6 (0, y, z) ≥ 0 for
all x, y, z ≥ 0, the conclusion follows by Theorem 1.
Case 2: k ∈ [−1, 1]. Since A ≥ 0, we apply Corollary 1 for
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1, namely f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈ [0, 4], is
satisfied if
9(k + 2)(x 2 + k x + 1)[(3 − 2k)x + 11 − k − 2k2 ] ≥ 2(k + 1)(1 − k)3 x(x − 1)2 .
Since
9 > 8 ≥ 2(1 − k)2 ,
k + 2 > k + 1,
x + k x + 1 ≥ x 2 − x + 1 ≥ (x − 1)2 ,
2
Since
k + 2 > k + 1,
x + k x + 1 ≥ x 2 − x + 1 ≥ x,
2
2x − y − z
≥ 0,
y 2 − yz + z 2
6x − y − z 18
≥ ,
y 2 + z2 x + y +z
x 1
≥ ,
4 y 2 + yz + z 2 x + y +z
2x + 3 y + 3z 24
≥ ,
y 2 + yz + z 2 x + y +z
y +z 18
≥ .
2 y 2 + 3 yz + 2z 2 7(x + y + z)
P 4.31. If x, y, z are nonnegative real numbers, no two of which are zero, then
2x − 3 y − 3z 6
+ ≥ 0.
y 2 + 4 yz + z 2 x + y + z
(
P3 (x, y, z) = (4 yz − x 2 ),
that is
A1 = P3 (1, 1, 1) = 27.
Therefore, f6 (x, y, z) has the highest coefficient
A = 6A1 = 162.
We have
f (x, 1, 1)
= 2(x − 3)(x 2 + 4x + 1) − 12(3x + 1) = 2(x 3 + x 2 − 29x − 9),
x 2 + 4x + 1
f6 (x, 1, 1) = (x + 2) f (x, 1, 1) + 36(x 2 + 4x + 1)2
= 2(x 2 + 4x + 1) (x + 2)(x 3 + x 2 − 29x − 9) + 18(x 2 + 4x + 1)
= 2(x 2 + 4x + 1)x(x 3 + 3x 2 − 9x + 5)
= 2x(x − 1)2 (x + 5)(x 2 + 4x + 1) ≥ 0.
Also, we have
f6 (0, y, z) ≥ 0
for y, z ≥ 0. This is true if the original inequality holds for x = 0. Thus, we need
to show that
−3( y + z) 2 y − 3z 2z − 3 y 6
+ + + ≥ 0,
y 2 + 4 yz + z 2 z2 y2 y +z
which is equivalent to
−3( y + z) 2( y 3 + z 3 ) − 3 yz( y + z) 6
+ + ≥ 0,
y2 + z + 4 yz
2 y 2z2 y +z
−3 2( y + z)2 − 9 yz 6
+ + ≥ 0,
( y + z) + 2 yz
2 y 2z2 ( y + z)2
−3 yz 2( y + z)2 − 9 yz 6 yz
+ + ≥ 0.
( y + z) + 2 yz
2 yz ( y + z)2
Highest Coefficient Cancellation Method for Nonnegative Variables 451
−3 6
+ 2t − 9 + ≥ 0,
t +2 t
2t 3 − 5t 2 − 15t + 12 ≥ 0,
(t − 4)(2t 2 + 3t − 3) ≥ 0.
First Solution. We apply Theorem 3. The conditions (a) and (c) are satisfied. The
condition (b) is satisfied if
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
4Ax(x − 1)3
f6 (x, 1, 1) − = 2x(x − 1)2 f (x),
27
where
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
2(x − 1)4 x 2
Af0,−2 (x) = ,
9
f6 (x, 1, 1) − Af0,−2 (x) = 2x(x − 1)2 g(x),
where
g(x) = (x 2 + 4x + 1)(x + 5) − x(x − 1)2 .
Since x 2 + 4x + 1 ≥ (x − 1)2 and x + 5 > x, it follows that g(x) > 0.
The condition (b) in Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Ax 2 for x ≥ 4. We
have
f6 (x, 1, 1) − Ax 2 = 2x g(x),
452 Vasile Cîrtoaje
where
89 − 1
k≥ ≈ 2.1085,
4
then (4k − 6)x + (1 − 4k)( y + z) 36(k + 1)
+ ≥ 0,
y 2 + k yz + z 2 (k + 2)(x + y + z)
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1, namely f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈ [0, 4], is
satisfied if
9(k + 2)(x 2 + k x + 1)[(2k − 3)x + 2k2 + k − 11] ≥ 2(k + 1)(k − 1)3 x(x − 1)2 .
Since
9 ≥ (x − 1)2 ,
k + 2 > k + 1,
x + k x + 1 > k x,
2
Indeed,
k(2k2 + 9k − 23) − 2(k − 1)3 = 15k2 − 29k + 2 > 15k2 − 30k = 15k(k − 2) > 0.
(k + 2)(x 2 + k x + 1)(x − 1)2 [(2k − 3)x + 2k2 + k − 11] ≥ 18(k + 1)(k − 1)3 x.
Since
k + 2 > k + 1,
x + k x + 1 > k x,
2
(x − 1)2 [(2k − 3)x + 2k2 + k − 11] ≥ 9[4(3 − 2k) + 11 − k − 2k2 ] = 9(2k2 + 9k − 23),
it suffices to show that
Indeed,
k(2k2 + 9k − 23) − 2(k − 1)3 = 15k2 − 29k + 2 > 15k2 − 30k = 15k(k − 2) > 0.
(t − 2)(B1 t 2 + B2 t + B3 ) ≥ 0,
where
y 2 + z2
t= ≥ 2,
yz
B1 = (k + 2)(4k − 6), B2 = (k + 2)(4k2 + 2k − 17), B3 = 8k3 − 5k2 − 25k − 2.
Since t ≥ 2, B1 > 0 and B2 ≥ 0, we have
P 4.32. If x, y, z are nonnegative real numbers, no two of which are zero, then
7x + 4 y + 4z 27
≥ .
4x 2 + yz x + y +z
(Vasile C., 2014)
454 Vasile Cîrtoaje
A1 = P3 (1, 1, 1) = 125.
A = −27A1 < 0.
Since A < 0, we only need to show that f6 (x, 1, 1) ≥ 0 and f6 (0, y, z) ≥ 0 for
x, y, z ≥ 0. The first inequality is true if the original inequality holds for y = z = 1.
So, we need to show that
7x + 8 2(4x + 11) 27
+ ≥ ,
4x 2 + 1 x +4 x +2
which is equivalent to
4( y + z) 4( y 3 + z 3 ) + 7 yz( y + z) 27
+ ≥ ,
yz 4 y 2z2 y +z
4 4( y 2 + z 2 ) + 3 yz 27
+ ≥ ,
yz 4 y 2z2 ( y + z)2
4( y + z)2 − 5 yz 27 yz
4+ ≥ .
4 yz ( y + z)2
Substituting
( y + z)2
t= , t ≥ 4,
yz
Highest Coefficient Cancellation Method for Nonnegative Variables 455
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
A1 = P3 (1, 1, 1) = 153 .
A = 3A1 = 3 · 153 .
We have
f (x, 1, 1)
= (9x − 4)(8x + 7) + 2(2x − 7)(7x 2 + 8)
8x + 7
= −(28x 3 − 170x 2 + x − 84),
Also, we have
f6 (0, y, z) ≥ 0
456 Vasile Cîrtoaje
for y, z ≥ 0. This is true if the original inequality holds for x = 0. Thus, we need
to show that
y + z 2z − 9 y 2 y − 9z 3
+ + + ≥ 0,
4 yz 7 y2 7z 2 y +z
which is equivalent to
y + z 2( y 3 + z 3 ) − 9 yz( y + z) 3
+ + ≥ 0,
4 yz 7 y 2z2 y +z
1 2( y 2 + z 2 ) − 11 yz 3
+ + ≥ 0,
4 yz 7 y 2z2 ( y + z)2
1 2( y + z)2 − 15 yz 3 yz
+ + ≥ 0,
4 7 yz ( y + z)2
Using the substitution
( y + z)2
t= , t ≥ 4,
yz
the inequality can be written as follows:
1 2t − 15 3
+ + ≥ 0,
4 7 t
8t 2 − 53t + 84 ≥ 0,
(t − 4)(8t − 21) ≥ 0.
First Solution. We apply Theorem 3. The conditions (a) and (c) are satisfied. The
condition (b) is satisfied if
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
4Ax(x − 1)3
f6 (x, 1, 1) − = 2x(x − 1)2 f (x),
27
where
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Second Solution. We apply Corollary 1 for
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
Highest Coefficient Cancellation Method for Nonnegative Variables 457
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
Af0,−2 (x) = 125(x − 1)4 x 2 ,
f6 (x, 1, 1) − Af0,−2 (x) = x(x − 1)2 g(x),
where
Since
7x 2 + y 2 + z 2 = 6x 2 + p2 − 2q,
+
the product (x 2 +4 y 2 +4z 2 ) has the same highest coefficient A1 as (6x 2 )(6 y 2 )(6z 2 ),
that is A1 = 216. Therefore, f6 (x, y, z) has the highest coefficient
A = −2A1 = −432.
458 Vasile Cîrtoaje
A = −3A1 = 81.
We have
f (x, 1, 1)
= (7x − 4)(4x 2 + 5) + 2(5 − 2x)(x 2 + 8)
4x 2 + 5
= 3(8x 3 − 2x 2 + x + 20),
Also, we have
f6 (0, y, z) ≥ 0
for y, z ≥ 0. This is true if the original inequality holds for x = 0. Thus, we need
to show that
−( y + z) 7 y − 2z 7z − 2 y 3
+ + ≥ ,
2( y 2 + z 2 ) y 2 + 4z 2 z 2 + 4 y 2 y +z
which is equivalent to
−( y + z) 26( y 3 + z 3 ) − yz( y + z) 3
+ ≥ ,
2( y + z )
2 2 4( y 4 + z 4 ) + 17 y 2 z 2 y +z
−1 26( y 2 + z 2 ) − 27 yz 3
+ ≥ .
2( y + z ) 4( y 2 + z 2 )2 + 9 y 2 z 2
2 2 ( y + z)2
If yz = 0, then the inequality is an equality. For yz = 0, write the inequality as
− yz 26 yz( y 2 + z 2 ) − 27 y 2 z 2 3 yz
+ ≥ 2 .
2( y + z )
2 2 4( y 2 + z 2 )2 + 9 y 2 z 2 y + z 2 + 2 yz
Using the substitution
y 2 + z2
t= , t ≥ 2,
yz
460 Vasile Cîrtoaje
First Solution. We apply Theorem 3. The conditions (a) and (c) are satisfied. The
condition (b) is satisfied if
4Ax(x − 1)3
f6 (x, 1, 1) ≥
27
for x ≥ 1. We have
4Ax(x − 1)3
f6 (x, 1, 1) − = 6x(x − 1)2 f (x),
27
where
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Second Solution. We apply Corollary 1 for
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
Af0,−2 (x) = (x − 1)4 x 2 ,
f6 (x, 1, 1) − Af0,−2 (x) = x(x − 1)2 g(x),
where
g(x) = 6(4x 2 + 5)(2x + 11) − x(x − 1)2 .
We have g(x) > 0 since 4x 2 + 5 > (x − 1)2 and 2x + 11 > x.
The condition (b) in Corollary 1 is satisfied if f6 (x, 1, 1) ≥ Ax 2 for x ≥ 4. We
have
f6 (x, 1, 1) − Ax 2 = 3x g(x),
where
g(x) = 2(4x 2 + 5)(x − 1)2 (2x + 11) − 27x > 0.
P 4.36. If x, y, z are nonnegative real numbers, no two of which are zero, then
2x 2 + yz 9 31(x − y)2 ( y − z)2 (z − x)2
≥ + .
y +z
2 2 2 2(x 2 + y 2 )( y 2 + z 2 )(z 2 + x 2 )
Since
y 2 + z 2 = −x 2 + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as
2 (2x 2 + yz)(−z 2 )(− y 2 ) − 9(−x 2 )(− y 2 )(−z 2 ) − 31(x − y)2 ( y − z)2 (z − x)2 ,
that is
A = 2(6 + 3) + 9 − 31(−27) = 32 · 27.
Also,
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
32(x − 1)4 x 2
Af0,−2 (x) = ,
3
4x(x − 1)2 g(x)
f6 (x, 1, 1) − Af0,−2 (x) = ,
3
where
g(x) = 3(x 2 + 1)(x + 2) − 8x(x − 1)2 .
For 0 ≤ x ≤ 1, we have x + 2 ≥ 3x, hence
For 1 ≤ x ≤ 4, we have
hence
g(x) > x[3x 2 + 6x − 8(x − 1)2 ] = x(5x − 2)(4 − x) ≥ 0.
where
g(x) = (x 2 + 1)(x − 1)2 (x + 2) − 216x.
The condition (c) in Corollary 1 is satisfied if the original inequality holds for
x = 0. Thus, we need to show that
yz y 2 z2 9 31( y − z)2
+2 + 2 ≥ + .
y 2 + z2 z2 y 2 2( y 2 + z 2 )
1 9 31(t − 2)
+ 2(t 2 − 2) ≥ + ,
t 2 2t
(t − 2)2 (t + 4) ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
P 4.37. If x, y, z are nonnegative real numbers, no two of which are zero, then
2x 2 − yz 9(x − y)2 ( y − z)2 (z − x)2
≥3+ 2 .
y − yz + z
2 2 (x − x y + y 2 )( y 2 − yz + z 2 )(z 2 − z x + x 2 )
Since
y 2 − yz + z 2 = −x 2 − yz + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as
where
(
P2 (x, y, z) = (2x 2 − yz)(−z 2 − x y)(− y 2 − xz), P3 (x, y, z) = (−x 2 − yz),
that is
Also,
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
31(x − 1)4 x 2
Af0,−2 (x) = ,
9
x(x − 1)2 g(x)
f6 (x, 1, 1) − Af0,−2 (x) = ,
9
where
g(x) = 18(x 2 − x + 1)(x + 1) − 31x(x − 1)2 .
For 0 ≤ x ≤ 1, we have x + 1 ≥ 2x, hence
For 1 ≤ x ≤ 4, we have
hence
g(x) > x[18x 2 − 31(x − 1)2 ] ≥ x[18x 2 − 32(x − 1)2 ] = 2x(7x − 4)(4 − x) ≥ 0.
464 Vasile Cîrtoaje
The condition (c) in Corollary 1 is satisfied if the original inequality holds for
x = 0. Thus, we need to show that
2
− yz y z2 9( y − z)2
+ 2 + ≥3+ 2 .
y − yz + z
2 2 z 2 y 2 y − yz + z 2
Using the substitution
y 2 + z2
t= , t ≥ 2,
yz
we may write the inequality as
−1 9(t − 2)
+ 2(t 2 − 2) ≥ 3 + ,
t −1 t −1
(t − 2)2 (t + 3) ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
−2 < k ≤ 2,
then
2x 2 + (2k + 1) yz 3(2k + 3) (8k2 + 30k + 31)(x − y)2 ( y − z)2 (z − x)2
≥ + + .
y2 + k yz + z2 k+2 (k + 2) ( y 2 + k yz + z 2 )
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
For
f6 (x, y, z) = (k + 2) [2x 2 + (2k + 1) yz](x 2 + k x y + y 2 )(x 2 + k xz + z 2 )
Highest Coefficient Cancellation Method for Nonnegative Variables 465
(
−3(2k + 3) ( y 2 + k yz + z 2 ) − (8k2 + 30k + 31)(x − y)2 ( y − z)2 (z − x)2 ,
we have
A = 9(k + 2)(2k2 + 19k + 48),
f6 (x, 1, 1) = 2(k + 2)x(x 2 + k x + 1)(x − 1)2 (x + k + 2).
P 4.38. If x, y, z are nonnegative real numbers, no two of which are zero, then
x y − yz + z x 3 5(x − y)2 ( y − z)2 (z − x)2
≥ + .
y 2 + z2 2 2(x 2 + y 2 )( y 2 + z 2 )(z 2 + x 2 )
Since
x y − yz + z x = −2 yz + q, y 2 + z 2 = −x 2 + p2 − 2q,
f6 (x, y, z) has the same highest coefficient A as
2 (−2 yz)(−z 2 )(− y 2 ) − 3(−x 2 )(− y 2 )(−z 2 ) − 5(x − y)2 ( y − z)2 (z − x)2 ,
that is
A = −12 + 3 − 5(−27) = 126.
Also,
Thus, we may apply Corollary 1 (for Eα,β = f0,−2 ) or Theorem 3. Since the last
method is more simple, we will apply it.
466 Vasile Cîrtoaje
The condition (c) in Theorem 3 is satisfied if the original inequality holds for
x = 0. Thus, we need to show that
− yz y 2 + z2 3 5( y − z)2
+ ≥ + .
y 2 + z2 yz 2 2( y 2 + z 2 )
−1 3 5(t − 2)
+t≥ + ,
t 2 2t
(t − 2)2 ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
P 4.39. If x, y, z are nonnegative real numbers, no two of which are zero, then
x y − 2 yz + z x 3(x − y)2 ( y − z)2 (z − x)2
≥ 2 .
y − yz + z
2 2 (x − x y + y 2 )( y 2 − yz + z 2 )(z 2 − z x + x 2 )
Since
x y − 2 yz + z xz = −3 yz + q, x 2 − x y + y 2 = −z 2 − x y + p2 − 2q,
that is
A = 3(−3)(−1 − 1)2 − 3(−27) = 45.
Also,
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 1 are satisfied if f6 (x, 1, 1) ≥ Af0,−2 (x) for x ∈
[0, 4]. We have
5(x − 1)4 x 2
Af0,−2 (x) = ,
9
x(x − 1)2 g(x)
f6 (x, 1, 1) − Af0,−2 (x) = ,
9
where
g(x) = 18(x 2 − x + 1) − 5x(x − 1)2 .
For 0 ≤ x ≤ 1, we have 18(x 2 − x + 1) ≥ 18(1 − x) ≥ 5(1 − x), hence
The condition (c) in Corollary 1 is satisfied if the original inequality holds for
x = 0. Thus, we need to show that
−2 yz y 2 + z2 3( y − z)2
+ ≥ 2 .
y 2 − yz + z 2 yz y − yz + z 2
Using the substitution
y 2 + z2
t= , t ≥ 2,
yz
we may write the inequality as
−2 3(t − 2)
+t≥ ,
t −1 t −1
(t − 2)2 ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
For
f6 (x, y, z) =(k + 2) [x( y + z) + (k − 1) yz](x 2 + k x y + y 2 )(x 2 + k xz + z 2 )
(
− 3(k + 1) ( y 2 + k yz + z 2 ) − (k2 + 3k + 5)(x − y)2 ( y − z)2 (z − x)2 ,
we have
A = 9(k + 2)(2k + 7),
f6 (x, 1, 1) = 2(k + 2)x(x 2 + k x + 1)(x − 1)2 .
P 4.40. If x, y, z are nonnegative real numbers, no two of which are zero, then
x + 3 y + 3z 7(x + y + z)
≥ .
( y + 2z)(2 y + z) 3(x y + yz + z x)
Since
2 y 2 + 5 yz + 2z 2 = 5 yz − 2x 2 + 2(p2 − 2q),
+
the product (2 y 2 + 5 yz + 2z 2 ) has the same highest coefficient A1 as
(
P3 (x, y, z) = (5 yz − 2x 2 ),
that is
A1 = P3 (1, 1, 1) = 27.
Therefore, f7 (x, y, z) ≥ 0 has the highest polynomial
f7 (x, 1, 1) ≥ 0
and
f7 (0, y, z) ≥ 0
for x, y, z ≥ 0. The first inequality is true if the original inequality holds for y =
z = 1. So, we need to show that
x +6 2(3x + 4) 7(x + 2)
+ ≥ ,
9 (x + 2)(2x + 1) 3(2x + 1)
which is equivalent to
x(x − 1)2 ≥ 0.
The second inequality is true if the original inequality holds for x = 0. Thus, we
need to prove that
3( y + z) y + 3z 3 y + z 7( y + z)
+ + ≥ ,
( y + 2z)(2 y + z) 2z 2 2 y2 3 yz
470 Vasile Cîrtoaje
which is equivalent to
3( y + z) ( y + z)3 7( y + z)
+ ≥ ,
( y + 2z)(2 y + z) 2 y 2z2 3 yz
3 ( y + z)2 7
+ ≥ ,
( y + 2z)(2 y + z) 2 y 2z2 3 yz
3 3( y 2 + z 2 ) − 8 yz
+ ≥ 0,
2( y 2 + z 2 ) + 5 yz 6 y 2z2
6( y 2 + z 2 )2 − ( y 2 + z 2 ) yz − 22 y 2 z 2 ≥ 0,
( y − z)2 [6( y 2 + z 2 ) + 11 yz] ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
P 4.41. If x, y, z are nonnegative real numbers, no two of which are zero, then
9x − 5 y − 5z 3(x + y + z)
+ ≥ 0.
2 y 2 − 3 yz + 2z 2 x y + yz + z x
Since
2 y 2 − 3 yz + 2z 2 = −3 yz − 2x 2 + 2(p2 − 2q),
+
the product (2 y 2 − 3 yz + 2z 2 ) has the same highest coefficient A1 as
(
P3 (x, y, z) = (−3 yz − 2x 2 ),
that is
A1 = P3 (1, 1, 1) = −125.
Therefore, f7 (x, y, z) ≥ 0 has the highest polynomial
which is equivalent to
−5 9 y 2 + 9z 2 − 14 yz 3
+ + ≥ 0,
2 y2 − 3 yz + 2z 2 2 y 2z2 yz
−5 9( y 2 + z 2 ) − 8 yz
+ ≥ 0,
2( y 2 + z 2 ) − 3 yz 2 y 2z2
18( y 2 + z 2 )2 − 43( y 2 + z 2 ) + 14 y 2 z 2 ≥ 0,
( y − z)2 (18 y 2 + 18z 2 − 7 yz) ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
1 − 17 1 + 17
k ∈ [a, −1] ∪ [1, b], a= ≈ −1.56155, b = ≈ 2.56155 ,
2 2
then (3 − k)x + (k − 1)( y + z) 3(k + 1) x + y +z
≥ · ,
y 2 + k yz + z 2 k+2 x y + yz + z x
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
For
(
f7 (x, y, z) = (k + 2)(x y + yz + z x) f (x, y, z) − 3(k + 1)(x + y + z) ( y 2 + k yz + z 2 ),
where
f (x, y, z) = [(3 − k)x + (k − 1)( y + z)](x 2 + k x y + y 2 )(x 2 + k xz + z 2 ),
472 Vasile Cîrtoaje
we have
A = −3(k + 1)(k − 1)3 p ≤ 0,
f7 (x, 1, 1) = 2(k + 2)x(x − 1)2 (x 2 + k x + 1)[(3 − k)x − k2 + k + 4] ≥ 0,
f7 (0, y, z) = yz( y + z)( y − z)2 [(k + 2)(3 − k)( y − z)2 + (13 + 7k − k2 − k3 ) yz] ≥ 0.
Observation 2. For k = 2, the inequality from Observation 1 turns into the well-
known Iran inequality:
1 9
≥ .
( y + z)2 4(x y + yz + z x)
P 4.42. If x, y, z are nonnegative real numbers, no two of which are zero, then
3x − y − z 3(x + y + z)
≥ .
y 2 + z2 2(x y + yz + z x)
We have
f7 (x, 1, 1) = 2(2x + 1) (3x − 2)(x 2 + 1)2 + 4(2 − x)(x 2 + 1) − 6(x + 2)(x 2 + 1)2
= 4x(x − 1)2 (x 2 + 1)(3x + 4).
− y − z 3 y − z 3z − y 3( y + z)
+ + ≥ ,
y 2 + z2 z2 y2 2 yz
− yz 3( y 2 + z 2 ) − 4 yz 3
+ ≥ .
y +z
2 2 yz 2
Highest Coefficient Cancellation Method for Nonnegative Variables 473
First Solution. Apply Theorem 6. The conditions (a) and (c) are satisfied. In what
concerns the condition (b), we have
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Second Solution. Apply Corollary 3 for
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Condition (a). It suffices to show that f7 (x, 1, 1) ≥ A(x + 2, 2x + 1) f0,−2 (x) for
0 ≤ x ≤ 4. We have
x 2 (x − 1)4 (x + 2)
A(x + 2, 2x + 1) f0,−2 (x) = ,
27
x(x − 1)2 f (x)
f7 (x, 1, 1) − A(x + 2, 2x + 1) f0,−2 (x) = ,
27
f (x) = 108(x 2 + 1)(3x + 4) − x(x − 1)2 (x + 2).
Since x 2 + 1 ≥ (x − 1)2 and 3x + 4 > 3x, we get
Condition (c). This condition is satisfied because the original inequality holds for
x = 0.
1 − 17 1 + 17
k ∈ [a, b], a= ≈ −1.56155, b = ≈ 2.56155 ,
2 2
then (3 − k)x + (k − 1)( y + z) 3(k + 1) x + y +z
≥ · .
y 2 + k yz + z 2 k+2 x y + yz + z x
(Vasile C., 2014)
Solution. Write the inequality as f7 (x, y, z) ≥ 0, where
(
f7 (x, y, z) = (k + 2)(x y + yz + z x) f (x, y, z) − 3(k + 1)(x + y + z) ( y 2 + k yz + z 2 ),
f (x, y, z) = [(3 − k)x + (k − 1)( y + z)](x 2 + k x y + y 2 )(x 2 + k xz + z 2 ).
Since
y 2 + k yz + z 2 = −x 2 + k yz + p2 − 2q,
+
the product ( y + k yz + z 2 ) has the same highest coefficient A1 as
2
(
P3 (x, y, z) = (−x 2 + k yz),
that is
A1 = P3 (x1, 1, 1) = (k − 1)3 .
Therefore, f7 (x, y, z) has the highest polynomial
We have
f (x, 1, 1)
= [(3 − k)x + 2k − 2](x 2 + k x + 1) + 2(k + 2)[(k − 1)x + 2]
x2 + kx + 1
= (3 − k)x 3 − (k2 − 5k + 2)x 2 + (2k2 − 3k + 3)x + 2k − 2,
Highest Coefficient Cancellation Method for Nonnegative Variables 475
f (0, y, z)
= (k − 1) y 2 z 2 + ( y 2 + z 2 + k yz) (3 − k)( y 2 + z 2 ) + (2k − 4) yz
y +z
= (3 − k)( y 2 + z 2 )2 − (k2 − 5k + 4)( y 2 + z 2 ) yz + (2k2 − 3k − 1) y 2 z 2 ,
f7 (0, y, z) f (0, y, z)
=(k + 2) − 3(k + 1) yz( y 2 + z 2 + k yz)
yz( y + z) y +z
=(k + 2)(3 − k)( y 2 + z 2 )2 − (k + 2)(k2 − 5k + 4)( y 2 + z 2 ) yz
+ (k + 2)(2k2 − 3k − 1) y 2 z 2 − 3(k + 1) yz( y 2 + z 2 + k yz)
=(k + 2)(3 − k)( y 2 + z 2 )2 − (k3 − 3k2 − 3k + 11)( y 2 + z 2 ) yz
+ 2(k3 − k2 − 5k − 1) y 2 z 2
=( y 2 + z 2 − 2 yz) (k + 2)(3 − k)( y 2 + z 2 ) − (k3 − k2 − 5k − 1) y 2 z 2
=( y − z)2 (k + 2)(3 − k)( y − z)2 + (13 + 7k − k2 − k3 ) yz ≥ 0.
Case 1: k ∈ [a, −1]∪[1, b]. Since A(p, q) ≤ 0, it suffices to show that f7 (x, 1, 1) ≥ 0
and f7 (0, y, z) ≥ 0 for x, y, z ≥ 0 (see Corollary 2). These condition are satisfied.
Case 2: k ∈ [−1, 1]. Since A(p, q) ≥ 0, we apply Corollary 3 for
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Condition (a). It suffices to show that f7 (x, 1, 1) ≥ A(x + 2, 2x + 1) f0,−2 (x) for
0 ≤ x ≤ 4, where
We have
(1 + k)(1 − k)3 x 2 (x − 1)4 (x + 2)
A(x + 2, 2x + 1) f0,−2 (x) = ,
27
476 Vasile Cîrtoaje
x + k x + 1 ≥ (x − 1)2 ,
2
6 ≥ x + 2,
it suffices to show that
9 ≥ (1 − k)3 .
Actually, we have 8 ≥ (1 − k)3 , which is equivalent to 2 ≥ 1 − k.
2(k + 2)(x 2 + k x + 1)(x − 1)2 [(3 − k)x − k2 + k + 4] ≥ 3(1 + k)(1 − k)3 (x + 2)x.
Since
k + 2 > k + 1,
x + k x + 1 > x + 2,
2
(x − 1)2 ≥ 9,
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Highest Coefficient Cancellation Method for Nonnegative Variables 477
P 4.44. If x, y, z are nonnegative real numbers, no two of which are zero, then
x + 13 y + 13z 27(x + y + z)
≥ .
y 2 + 4 yz + z 2 2(x y + yz + z x)
(Vasile C., 2014)
(
P3 (x, y, z) = (−x 2 + 4 yz),
that is
A1 = P3 (1, 1, 1) = 27.
Therefore, f7 (x, y, z) has the highest polynomial
Since A(p, q) ≤ 0, according to Corollary 2, we only need to show that the original
inequality holds for y = z = 1, and for x = 0. For y = z = 1, the original inequality
becomes
x + 26 2(13x + 14) 27(x + 2)
+ 2 ≥ ,
6 x + 4x + 1 2(2x + 1)
x 4 − 10x 3 + 33x 2 − 40x + 16 ≥ 0,
(x − 1)2 (x − 4)2 ≥ 0.
Also, for x = 0, the original inequality becomes
where
y 2 + z2
t= .
yz
The equality occurs for x = y = z, and for x/4 = y = z (or any cyclic permutation).
P 4.45. If x, y, z are nonnegative real numbers, no two of which are zero, then
−x + y + z x + y +z
≥ .
2x 2 + yz x y + yz + z x
Since (
P3 (x, y, z) = (2x 2 + yz)
has the highest coefficient
A1 = P3 (1, 1, 1) = 27,
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Highest Coefficient Cancellation Method for Nonnegative Variables 479
with equality for x = y = z, and for x = 0 and y = z (or any cyclic permutation).
2x − y − z 2 y − z − x 2z − x − y
+ + 2 ≤ 0.
x 2 + yz y2 + z x z +xy
P 4.46. If x, y, z are nonnegative real numbers, no two of which are zero, then
11x − 3 y − 3z 3(x + y + z)
≤ .
2x 2 + 3 yz x y + yz + z x
A1 = P3 (1, 1, 1) = 125.
We have
f (x, 1, 1)
= (11x − 6)(3x + 2) + 2(8 − 3x)(2x 2 + 3)
3x + 2
= −(12x 3 − 65x 2 + 14x − 36),
480 Vasile Cîrtoaje
− y − z 11 y − 3z 11z − 3 y 3( y + z)
+ + ≤ ,
yz 2 y2 2z 2 yz
which is equivalent to
4( y + z) 3( y 3 + z 3 ) − 11 yz( y + z)
+ ≥ 0,
yz 2 y 2z2
( y + z)( y − z)2 ≥ 0.
First Solution. Apply Theorem 6. The conditions (a) and (c) are satisfied. In what
concerns the condition (b), we have
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Second Solution. Apply Corollary 3 for
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 3 are satisfied if f7 (x, 1, 1) ≥ A(x+2, 2x+1) f0,−2 (x)
for x ∈ [0, 4]. We have
P 4.47. If x, y, z are nonnegative real numbers, no two of which are zero, then
1 1 1 1 2
+ + ≥ + .
2x 2 + yz 2 y 2 + z x 2z 2 + x y x y + yz + z x x 2 + y 2 + z 2
A(p, q) = −A1 p2 ,
Since
A1 = P3 (1, 1, 1) = 27,
we have
A(p, q) = −27p2 < 0.
According to Corollary 2, we only need to prove the original inequality for y = z = 1
and for x = 0. For y = z = 1, the original inequality becomes
1 2 1 2
+ ≥ + ,
2x 2 + 1 x + 2 2x + 1 x 2 + 2
which is equivalent to
2x(x − 1) 2x(x − 1)
≥ ,
(x + 2)(x 2 + 2) (2x + 1)(2x 2 + 1)
1 1 2
+ ≥ 2 ,
2 y 2 2z 2 y + z2
which is equivalent to
( y 2 − z 2 )2 ≥ 0.
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
(a) If k ≥ 2, then
1 4(k + 1) 5k − 4
k(k + 1) ≥ + ;
x 2 + k yz x y + yz + z x x 2 + y 2 + z 2
1
(b) If ≤ k ≤ 2, then
8
1 4k2 − 3k + 2 2(−2k2 + 6k − 1)
k(k + 1) ≥ + .
x2 + k yz x y + yz + z x x 2 + y 2 + z2
P 4.48. If x, y, z are nonnegative real numbers, no two of which are zero, then
x( y + z) y(z + x) z(x + y) x 2 + y 2 + z2
+ 2 + 2 ≤ .
x + 5 yz y + 5z x z + 5x y
2 x y + yz + z x
(Vasile C., 2012)
and
P2 (x, y, z) = yz( y 2 + 5z x)(z 2 + 5x y),
respectively. Therefore, we have
A(p, q) = P3 (1, 1, 1)(p2 − 2q) + P2 (1, 1, 1)q = 63 (p2 − 2q) + 3 · 62 q = 108(2p2 − 3q).
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Condition (a). It suffices to show that f8 (x, 1, 1)(x) ≥ A(x + 2, 2x + 1) f0,−2 (x) for
0 ≤ x ≤ 4, where
A(x + 2, 2x + 1) = 108(2x 2 + 2x + 5).
We have
4x 2 (x − 1)4 (2x 2 + 2x + 5)
A(x + 2, 2x + 1) f0,−2 (x) = ,
3
x(x − 1)2 f (x)
f8 (x, 1, 1)(x) − A(x + 2, 2x + 1) f0,−2 (x) = ,
3
where
Condition (b). It suffices to show that f8 (x, 1, 1)(x) ≥ A(x + 2, 2x + 1)x 2 for
x ≥ 4. The inequality is equivalent to
5x + 1 > 5x,
5
5x 2 + 7x + 18 > (2x 2 + 2x + 5),
2
Highest Coefficient Cancellation Method for Nonnegative Variables 485
z y y 2 + z2
+ ≤ ,
y z yz
which is an identity.
The equality occurs for x = y = z, and also for x = 0 (or any cyclic permutation).
x( y + z) y(z + x) z(x + y) x 2 + y 2 + z2
+ + ≤ 1 + ,
x 2 + 2 yz y 2 + 2z x z 2 + 2x y x y + yz + z x
x( y + z) y(z + x) z(x + y) 1 x 2 + y 2 + z2
+ + ≤ + ,
x 2 + 4 yz y 2 + 4z x z 2 + 4x y 5 x y + yz + z x
x( y + z) y(z + x) z(x + y) x 2 + y 2 + z2
+ + ≤ .
x 2 + 5 yz y 2 + 5z x z 2 + 5x y x y + yz + z x
P 4.49. If x, y, z are nonnegative real numbers, no two of which are zero, then
15q − 2p2 ≥ 0.
and (
P3 (x, y, z) = (x 2 + yz),
respectively. Therefore, we have
A(p, q) = −P2 (1, 1, 1)p2 − P3 (1, 1, 1)(15q − 2p2 ) = −12p2 − 8(15q − 2p2 ) < 0.
2x 15(2x + 1)
+4≥ ,
x2 + 1 (x + 2)2
which is equivalent to
(x − 1)2 (2x − 1)2 ≥ 0.
which is equivalent to
( y + z)4 ≥ 15 y 2 z 2 ,
( y + z)4 − 16 y 2 z 2 + y 2 z 2 ≥ 0,
( y − z)2 [( y + z)2 + 4 yz] + y 2 z 2 ≥ 0.
x
The equality occurs for x = y = z, and for = y = z (or any cyclic permutation).
2
Highest Coefficient Cancellation Method for Nonnegative Variables 487
P 4.50. If x, y, z are nonnegative real numbers, no two of which are zero, then
x( y + z) y(z + x) z(x + y) x y + yz + z x
+ + ≥1+ 2 .
x 2 + 2 yz y 2 + 2z x z 2 + 2x y x + y 2 + z2
and (
P3 (x, y, z) = (x 2 + 2 yz),
respectively. Therefore, we have
2x 2(x + 1) 2x + 1
+ ≥1+ 2 ,
x2 + 2 2x + 1 x +2
which is equivalent to
(x − 1)2 ≥ 0.
which is equivalent to
(a) If 0 ≤ k ≤ 2, then
k x( y + z) 2k − 1 x y + yz + z x
≥ + 2 ;
2 x 2 + k yz k+1 x + y 2 + z2
(b) If k ≥ 2, then
k x( y + z) 2k − 1 x y + yz + z x
≥1+ · 2 .
2 x 2 + k yz k+1 x + y 2 + z2
x( y + z) y(z + x) z(x + y) 1 7 x y + yz + z x
+ + ≥ + · ,
x 2 + 4 yz y 2 + 4z x z 2 + 4x y 2 10 x 2 + y 2 + z 2
x( y + z) y(z + x) z(x + y) x y + yz + z x
+ + ≥1+ 2 ,
x 2 + 2 yz y 2 + 2z x z 2 + 2x y x + y 2 + z2
x( y + z) y(z + x) z(x + y) (x + y + z)2
+ 2 + 2 ≥ 2 ,
x 2 + yz y + zx z +xy x + y 2 + z2
x( y + z) y(z + x) z(x + y) 2(x y + yz + z x)
+ + ≥ .
2x 2 + yz 2 y 2 + z x 2z 2 + x y x 2 + y 2 + z2
that is
A(p, q) = −5p2 + 24q.
We have
f8 (x, 1, 1) = 2(2x + 1)2 [(x 2 + 1)2 + 4(x 2 + 1)] + 2(5x 2 − 28x − 4)(x 2 + 1)2
= 2(x 2 + 1)(9x 4 − 24x 3 + 22x 2 − 8x + 1)
= 2(x 2 + 1)(x − 1)2 (3x − 1)2 ,
The condition (a) in Corollary 3 are satisfied if f8 (x, 1, 1) ≥ A(x+2, 2x+1) f1/3,−2 (x)
for x ∈ [0, 4], where
We have
2(x − 1)2 (3x − 1)2 g(x)
f8 (x, 1, 1) − A(x + 2, 2x + 1) f1/3,−2 (x) = ,
3969
where
For the nontrivial case −5x 2 + 28x + 4 ≥ 0, since x 2 + 1 > x 2 and 2(x − 1)2 > 1, it
suffices to show that
(3x − 1)2 ≥ −5x 2 + 28x + 4.
490 Vasile Cîrtoaje
Indeed,
(3x − 1)2 − (−5x 2 + 28x + 4) = 14x 2 − 34x − 3 ≥ 56x − 34x − 3 = 22x − 3 > 0.
2x y 2 yz 2z x 30(x y + yz + z x)
+ + +7≥ .
x 2 + y 2 y 2 + z2 z2 + x 2 (x + y + z)2
that is
A(p, q) = 6p2 − (7p2 − 30q) = 30q − p2 .
We have
f8 (x, 1, 1) = 2(x + 2)2 [(x 2 + 1)2 + 4x(x 2 + 1)] + 2(7x 2 − 32x − 2)(x 2 + 1)2
= 4(x 2 + 1)(4x 4 − 12x 3 + 13x 2 − 6x + 1)
= 4(x 2 + 1)(x − 1)2 (2x − 1)2 ,
The condition (a) in Corollary 3 are satisfied if f8 (x, 1, 1) ≥ A(x+2, 2x+1) f1/2,−2 (x)
for x ∈ [0, 4], where
We have
4(x − 1)2 (2x − 1)2 g(x)
f8 (x, 1, 1) − A(x + 2, 2x + 1) f1/2,−2 (x) = ,
2025
where
g(x) = 2025(x 2 + 1) − (−x 2 + 56x + 26)(x − 1)2
≥ 2025(x − 1)2 − (−x 2 + 56x + 26)(x − 1)2
≥ 234(x − 1)2 − (−x 2 + 56x + 26)(x − 1)2
= (x − 1)2 (4 − x)(52 − x) ≥ 0.
For the nontrivial case −x 2 + 56x + 26 ≥ 0, since x 2 + 1 > x 2 and (x − 1)2 > 2, it
suffices to show that
8(2x − 1)2 ≥ −x 2 + 56x + 26.
Indeed,
8(2x − 1)2 − (−x 2 + 56x + 26) = 33x 2 − 88x − 18 > 32x − 88x − 24
= 8(x − 3)(4x + 1) > 0.
xy 8(x y + yz + z x)
+7≥ .
x2 + x y + y2 (x + y + z)2
2x y 2 yz 2z x x 2 + y 2 + z2 5
+ + + ≥ .
(x + y) 2 ( y + z) 2 (z + x) 2 x y + yz + z x 2
Since ( y +z)2 = (p− x)2 = x 2 + p(p−2x), the polynomial of degree eight f8 (x, y, z)
has the same highest polynomial A(p, q) as
4q yz(z 2 )( y 2 ) + (2p2 − 9q)(z 2 )(x 2 )( y 2 ),
that is
A(p, q) = 12q − (2p2 − 9q) = 21q − 2p2 .
We have
f8 (x, 1, 1) = 4(2x + 1) (x + 1)4 + 8x(x + 1)2 + 4(2x 2 − 10x − 1)(x + 1)4
= 8x 2 (x + 1)2 (x − 1)2 ,
f8 (0, y, z) = 4 y 4 z 4 + 2( y + z)2 − 9 yz y 2 z 2 ( y 2 + z 2 )
= y 2 z 2 ( y − z)2 (2 y 2 + 2z 2 + 3 yz).
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 3 are satisfied if f8 (x, 1, 1) ≥ A(x+2, 2x+1) f0,−2 (x)
for x ∈ [0, 4], where
We have
(x − 1)2 x 2 g(x)
f8 (x, 1, 1) − A(x + 2, 2x + 1) f0,−2 (x) = ,
81
where
For the nontrivial case −2x 2 + 34x + 13 ≥ 0, since (x − 1)2 > 1, it suffices to show
that
8(x + 1)2 ≥ −2x 2 + 34x + 13.
Indeed,
Observation 1. In the case 2, we can give a more simple solution based on The-
orem 6. The conditions (a) and (c) are satisfied. Thus, it suffices to show that
4x(x − 1)3
f8 (x, 1, 1) ≥ A(x + 2, 2x + 1) for x ≥ 1. We have
27
2x
with equality for x = y = z, and also for
= y = z (or any cyclic
5 − 2k − 1
permutation).
2x y 2
x + y 2 + z2
− 3 + ( 5 + 1) − 1 ≥ 0;
x2 + y2 x y + yz + z x
5x y 8 x 2 + y 2 + z2
− 3 + − 1 ≥ 0;
2x 2 + x y + 2 y 2 3 x y + yz + z x
2
3x y 4 x + y 2 + z2
− 3 +
− 1 ≥ 0;
x2 + x y + y2 3 x y + yz + z x
4x y
x 2 + y 2 + z2
− 3 + 2 − 1 ≥ 0.
(x + y)2 x y + yz + z x
P 4.54. If x, y, z are nonnegative real numbers, no two of which are zero, then
2 2 2 8 1
+ + ≥ 2 + .
x 2 + y 2 y 2 + z2 z2 + x 2 x + y 2 + z 2 x y + yz + z x
that is
A(p, q) = p2 + 6q, A(x + 2, 2x + 1) = x 2 + 16x + 10.
Highest Coefficient Cancellation Method for Nonnegative Variables 495
We have
f8 (x, 1, 1) = 2(2x + 1)x 2 + 2) (x 2 + 1)2 + 4(x 2 + 1) − 2(x 2 + 16x + 10)(x 2 + 1)2
= 4x(x 2 + 1)(x 4 − x 2 − 2x + 2)
= 4x(x 2 + 1)(x − 1)2 (x 2 + 2x + 2).
which is equivalent to
( y − z)2 (2 y 2 + 2z 2 + 3 yz) ≥ 0.
First Solution. Apply Theorem 6. The conditions (a) and (c) are satisfied. In what
concerns the condition (b), we have
The equality occurs for x = y = z, and for x = 0 and y = z (or any cyclic
permutation).
Second Solution. Apply Corollary 3 for
(x − 1)4 x 2
Eα,β (x) = f0,−2 (x) = .
81
The condition (a) in Corollary 3 is satisfied if f8 (x, 1, 1) ≥ A(x +2, 2x +1) f0,−2 (x)
for x ∈ [0, 4]. We have
which is equivalent to
2x(2x − 3) ≥ 0.
The condition (c) in Corollary 3 is satisfied since the original inequality holds for
x = 0.
Observation. Similarly, we can prove the following generalization:
10 + 136
1− 2≤ k ≤ ,
9
then k+2 8 − 4k 4k + 1
≥ 2 + ,
x2 + kx y + y2 x + y 2 + z 2 x y + yz + z x
with equality for x = y = z, and also for x = 0 and y = z (or any cyclic permutation).
2x y 2 yz 2z x 4(x y + yz + z x)
+ + +1≥ .
x 2 + y 2 y 2 + z2 z2 + x 2 x 2 + y 2 + z2
Since
y 2 + z 2 = −x 2 + p2 − 2q,
the polynomial of degree eight f8 (x, y, z) has the same highest polynomial A(p, q)
as
2(p2 − 2q) yz(−z 2 )(− y 2 ) + (p2 − 6q)(−z 2 )(−x 2 )(− y 2 ),
that is
A(p, q) = 6(p2 − 2q) − (p2 − 6q) = 5p2 − 6q = 2(p2 − 3q) + 3p2 > 0,
A(x + 2, 2x + 1) = 5x 2 + 8x + 14 > 0.
For x = 0, the original inequality becomes
2 yz 4 yz
+1≥ 2 ,
y 2 + z2 y + z2
which is equivalent to
( y − z)2 ≥ 0.
Also, we have
f8 (x, 1, 1) = 2(x 2 + 2) (x 2 + 1)2 + 4x(x 2 + 1) + 2(x 2 − 8x − 2)(x 2 + 1)2
= 4x 2 (x 2 + 1)(x − 1)2 .
First Solution. Apply Theorem 6. The conditions (a) and (c) are satisfied. In what
concerns the condition (b), we have
For x ≥ 1, we get
g(x) > 27(x − 1)(x 2 + 1) − (x − 1)(5x 2 + 8x + 14)
= (x − 1)(27x 2 − 8x + 13) ≥ 0.
The equality holds for x = y = z, and also for x = 0 and y = z (or any cyclic
permutation).
Second Solution. Apply Corollary 3 for
x 2 (x − 1)4
Eα,β (x) = f0,−2 (x) = .
81
Condition (a). It suffices to show that f8 (x, 1, 1) ≥ A(x + 2, 2x + 1) f0,−2 (x) for
0 ≤ x ≤ 4. We have
x 2 (x − 1)4 (5x 2 + 8x + 14)
A(x + 2, 2x + 1) f0,−2 (x) = ,
81
x 2 (x − 1)2 f (x)
f8 (x, 1, 1) − A(x + 2, 2x + 1) f0,−2 (x) = ,
81
f (x) = 324(x + 1) − (x − 1) (5x + 8x + 14).
2 2 2
Condition (c). This condition is satisfied because the original inequality holds for
x = 0.
On Popoviciu’s Inequality
f (a1 ) + f (a2 ) + · · · + f (an ) + n(n − 2) f (a) ≥ (n − 1)[ f (b1 ) + f (b2 ) + · · · + f (bn )],
1! n
1 !
where a = a j and bi = a j for all i.
n j=1 n − 1 j =i
a1 ≤ · · · ≤ am ≤ a ≤ am+1 ≤ · · · ≤ an ,
b1 ≥ · · · ≥ bm ≥ a ≥ bm+1 ≥ · · · ≥ bn .
We can get the desired inequality by summing the inequalities
499
500 Vasile Cîrtoaje
b1 ≥ · · · ≥ bm ≥ a, c + (m − 1)a = b1 + · · · + bm ,
Proof. We use the induction method. For n = 2, the equality occurs. Suppose that
the inequality holds for n − 1 numbers, n ≥ 3, and show that it also holds for n
numbers. Let
a1 + a2 + · · · + an a1 + a2 + · · · + an−1
a= , b= .
n n−1
By the induction hypothesis,
ai + a j
(n − 3)[ f (a1 ) + f (a2 ) + · · · + f (an−1 )] + (n − 1) f (b) ≥ 2 f .
1≤i< j≤n−1
2
Since
f (a1 ) + f (a2 ) + · · · + f (an−1 ) ≥ (n − 1) f (b)
On Popoviciu’s Inequality 501
Since
n−1
ai + an
(n − 2)an + na = 2 ,
i=1
2
we will apply Karamata’s inequality.
Case 1: a1 ≤ a2 ≤ · · · ≤ an , 2a ≤ a1 + an . Since
a + a a + a an−1 + an an−1 + an
1 n 2 n
an ≥ max , ,..., =
2 2 2 2
and a + a a + a
1 n 2 n an−1 + an a1 + an
a ≤ min , ,..., = ,
2 2 2 2
the inequality (***) follows from Karamata’s inequality.
Case 2: a1 ≥ a2 ≥ · · · ≥ an , 2a ≥ a1 + an . Since
a + a a + a an−1 + an a1 + an
1 n 2 n
a ≥ max , ,..., = ,
2 2 2 2
a + a a + a an−1 + an an−1 + an
1 n 2 n
an ≤ min , ,..., = ,
2 2 2 2
the inequality (***) follows from Karamata’s inequality.
En (a1 , a2 , . . . , an ) ≥ n − 1,
where
f (a1 ) + f (a2 ) + · · · + f (an ) − n f (a)
En (a1 , a2 , . . . , an ) = .
f (b1 ) + f (b2 ) + · · · + f (bn ) − n f (a)
502 Vasile Cîrtoaje
For some convex functions, the minimum (greatest lower bound) of En is just n − 1,
but it is greater for other functions. Thus, for f (x) = x 2 , we have
1! n
1 !
where a = a j and bi = a j for all i. Also, for f (x) = x 3 , x ≥ 0, if
n j=1 n − 1 j =i
a1 , a2 , . . . , an are nonnegative numbers, then
which is the inequality in P 1.7 from Volume 4 (for k = 3), and the inequality in P
1.2 from Volume 5 (for n = 6.)
On Popoviciu’s Inequality 503
5.2 Applications
a1 a2 · · · an = 1,
then
1 1 1
a1n−1 + a2n−1 + · · · + ann−1 + n(n − 2) ≥ (n − 1) + + ··· + .
a1 a2 an
a1 a2 · · · an = 1,
then
n−1 1 1 1
a1n−1 + a2n−1 + · · ·+ ann−1 + n(n− 2) ≥ a1 + a2 + · · · + an + + + ··· + .
2 a1 a2 an
a1 a2 · · · an = 1,
then
(n − 1)(a12 + a22 + · · · + an2 ) + n ≥ (a1 + a2 + · · · + an )2 .
a b + bc + cd + d a = 4,
then
a b c d
1+ 1+ 1+ 1+ ≥ (a + b + c + d)2 .
b c d a
a1 + a2 + · · · + an = n,
then
1 − a1 1 − a2 1 − an
1+ 1+ ··· 1 + ≥ n−1 a1 a2 · · · an .
n−1 n−1 n−1
504 Vasile Cîrtoaje
a1 + a2 + · · · + an = 1,
then n
1 1 1 1
a1 + −2 a2 + − 2 · · · an + −2 ≥ n+ −2 .
a1 a2 an n
b1 b2 bn a1 a2 an
+ + ··· + ≥ + + ··· + ,
a1 a2 an b1 b2 bn
where
1
bi = aj
n − 1 j =i
for all i.
1 1 1
a1 + a2 + · · · + an = + + ··· + ,
a1 a2 an
then
1 1 1
(a) + + ··· + ≥ 1;
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an
1 1 1
(b) + + ··· + ≤ 1.
n − 1 + a1 n − 1 + a2 n − 1 + an
1 1 1
a 1 + a2 + · · · + a n = + + ··· + = ns,
a1 a2 an
then
1 1 1 1 1 1
+ +· · ·+ ≥ + +· · ·+ .
a1 + n − 1 a2 + n − 1 an + n − 1 1 + ns − a1 1 + ns − a2 1 + ns − an
On Popoviciu’s Inequality 505
where
1
a= (a1 + a2 + · · · + an ).
n
a1 a2 · · · an = 1,
then
a1 a2 an
2(a1n + a2n + · · · + ann ) + n(n − 2) ≥ n + + ··· + .
a2 a3 a1
a1 + a2 + · · · + an = n,
then a1 − a2 a2 − a3 a n − a1 2
1+ 1+ ··· 1 + ≥ (a1 a2 · · · an ) n .
n n n
506 Vasile Cîrtoaje
a1 + a2 + · · · + an = n,
then
2 1 1 1 1 1 1
+ + ··· + +n−2≥ + + ··· + .
n a1 a 2 an a1 − a2 a2 − a3 an − a1
1+ 1+ 1+
n n n
5.17. Let f be a convex function on (0, ∞). If a1 , a2 , . . . , an are positive real num-
bers, then
1 1 1
f a1 + + f a2 + + · · · + f an + ≥
a2 a3 a1
1 1 1
≥ f a1 + + f a2 + + · · · + f an + .
a1 a2 an
On Popoviciu’s Inequality 507
5.3 Solutions
Solution. We can get this inequality by adding the inequality in P 5.1 and the
inequality
ain−1 + n − 2 ≥ (n − 1)ai , i = 1, 2, . . . , n.
a1 a2 · · · an = 1,
then
(n − 1)(a12 + a22 + · · · + an2 ) + n ≥ (a1 + a2 + · · · + an )2 .
(F. Shleifer, 1979)
x 1 + x 2 + · · · + x n = 0.
Applying the inequality from Theorem 2 to the convex function f (x) = e x , we get
x i +x j
(n − 2) (e x 1 + e x 2 + · · · + e x n ) + n ≥ 2 e 2 .
1≤i< j≤n
x 1 = 2 ln a1 , x 2 = 2 ln a2 , . . . , x n = 2 ln an ,
Since
2 ai a j = (a1 + a2 + · · · + an )2 − (a12 + a22 + · · · + an2 ),
1≤i< j≤n
a1 = a2 = · · · = an = 1.
On Popoviciu’s Inequality 509
Solution. Let
a = a1 + a2 + · · · + an .
Since
bi a ai a
(n − 1)= − 1, = − n + 1, i = 1, 2, . . . , n,
ai ai bi bi
the inequality becomes
1 1 1 n−2 1 1 1
+ + ··· + + ≥ (n − 1) + + ··· + ,
a1 a2 an a b1 b2 bn
which is just Popoviciu’s inequality from Theorem 1 applied to the convex function
1
f (x) = , x > 0.
x
For n ≥ 3, the equality holds if and only if a1 = a2 = · · · = an .
Remark. We can also prove this inequality using the Cauchy-Schwarz inequality
n−1 1 1
! ≤ , i ∈ {1, 2, . . . , n}.
aj n − 1 j =i a j
j =i
Setting
1 1 1 1
a = a1 + a2 + · · · + an , = + + ··· + ,
A a1 a2 an
we can write this inequalities as
1 1 1 1
≤ − ,
bi n − 1 A ai
ai 1 ai
≤ −1 .
bi n−1 A
Therefore,
n
1 a 1 a
n n
a i bi
≤ −n = −1 = .
i=1
bi n−1 A n − 1 i=1 ai a
i=1 i
512 Vasile Cîrtoaje
1 1 1
a1 + a2 + · · · + an = + + ··· + ,
a1 a2 an
then
1 1 1
(a) + + ··· + ≥ 1;
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an
1 1 1
(b) + + ··· + ≤ 1.
n − 1 + a1 n − 1 + a2 n − 1 + an
(Vasile C., 1996)
Solution. (a) We use the contradiction method. For the sake of contradiction, as-
sume that
1 1 1
+ + ··· + < 1.
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an
1 − xi
ai = , i = 1, 2, . . . , n,
(n − 1)x i
we get
x 1 + x 2 + · · · + x n < 1,
which is equivalent to
1 − x i > (n − 1) yi ,
where
1
yi = x j, i = 1, 2, . . . , n.
n − 1 j =i
Therefore, we have
n
1 − xi n
yi
a1 + a2 + · · · + an = > .
i=1
(n − 1)x i x
i=1 i
In addition, since
n
x i (n − 1)x i
n
1 1 1
> = + + ··· + ,
y
i=1 i i=1
1 − xi a1 a2 an
On Popoviciu’s Inequality 513
we have
1 1 1
a1 + a2 + · · · + an > + + ··· + ,
a1 a2 an
which contradicts the hypothesis
1 1 1
a1 + a2 + · · · + an = + + ··· + .
a1 a2 an
1 1 1
a1 + a2 + · · · + an = + + ··· + = ns,
a1 a2 an
then
1 1 1 1 1 1
+ +· · ·+ ≥ + +· · ·+ .
a1 + n − 1 a2 + n − 1 an + n − 1 1 + ns − a1 1 + ns − a2 1 + ns − an
which leads to
s ≥ 1.
Applying Popoviciu’s inequality from Theorem 1 to the convex function
1
f (x) = , x > 0,
1 + (n − 1)x
we get
n
1 n(n − 2) n
1
+ ≥ (n − 1) .
i=1
1 + (n − 1)ai 1 + (n − 1)s i=1
1 + ns − ai
514 Vasile Cîrtoaje
which is equivalent to
n
1 1
≥ .
i=1 (ai + n − 1) 1
+n−1 1 + (n − 1)s
ai
1 1 1 n2 n
+ + ··· + ≥ = .
A 1 A2 An A1 + A 2 + · · · + A n 2(n − 1)s + n2 − 2n + 2
Consequently, it is enough to prove the inequality
n 1
≥ ,
2(n − 1)s + n2 − 2n + 2 1 + (n − 1)s
which reduces to s ≥ 1. For n ≥ 3, the equality holds if and only if a1 = · · · = an = 1.
f (x) = − x, x ≥ 0,
which is weaker that the desired inequality. To prove the desired inequality, con-
sider the nontrivial case where at least two of a, b, c are positive, and write it in the
homogeneous form
"
# "
#
10 3(a + b + c) − a − b − c ≤ 9 2(a + b) − a − b ,
or, equivalently,
2
10
2 a− b
a− b ≤9
.
3(a + b + c) + a + b + c 2(a + b) + a + b
This is true if
10 9
≤
,
3(a + b + c) + a+ b+ c 2(a + b) + a + b
which is equivalent to
3
9 − 10 2(a + b) ≥ a + b.
2
is for
k = ( 3 − 1)( 3 + 2) ≈ 2.303.
1 5− x 1 5− y 1 5−z
= ,
= ,
= ,
2 + 4 + 5a 30 2 + 4 + 5b 30 2 + 4 + 5c 30
1 5−u 1 5−v
= ,
= ,
2 + 4 + 5d 30 2 + 4 + 5e 30
which involves
16x 16 y 16z 16u 16v
a= , b= , c= , d= , e= ,
(5 − x)2 (5 − y)2 (5 − z)2 (5 − u)2 (5 − v)2
and
0 < x < 5, 0 < y < 5, 0 < z < 5, 0 < u < 5, 0 < v < 5,
we need to show that
x + y +z+u+v <5
implies
2 2 2 2 2
5− x 5− y 5−z 5−u 5−v
x yzuv < .
4 4 4 4 4
It is easy to see that if x increases, then the left side of this inequality increases,
while the right side decreases. Therefore, it suffices to show that
x + y +z+u+v =5
implies
2 2 2 2 2
5− x 5− y 5−z 5−u 5−v
≥ x yzuv.
4 4 4 4 4
f (x) = − ln x, x > 0,
gives
4 4 4 4 4 x + y + z + u + v 15
5− x 5− y 5−z 5−u 5−v
≥ x yzuv ,
4 4 4 4 4 5
On Popoviciu’s Inequality 517
4 4 4 4 4
5− x 5− y 5−z 5−u 5−v
≥ x yzuv.
4 4 4 4 4
Thus, it suffices to show that x yzuv ≤ 1. By the AM-GM inequality, we have
x + y + z + u + v 5
x yzuv ≤ = 1.
5
The equality holds for a = b = c = d = e = 1.
Remark. In the same manner, we can prove the following generalization:
• If a1 , a2 , . . . , an (n ≥ 3) are positive real numbers such that
a1 a2 · · · an = 1,
then
n
1 1
≤ ,
i=1 n−1+ (n − 1)2 + 4nai 2
with equality for a1 = a2 = · · · = an = 1.
Solution. We will apply the contradiction method. Assume that the reverse in-
equality holds, namely
1 1 1 n
+ + > ,
1 + pa1 1 + pa2 1 + pan 1+p
n − (n − 1)x
f (x) = ln
q−x
yields
[n − (n − 1)x 1 ][n − (n − 1)x 2 ] · · · [n − (n − 1)x n ] 1
≥ .
(q − x 1 )(q − x 2 ) · · · (q − x n ) (q − 1)n
Multiplying this inequality and (**) gives
(q − x 1 )(q − x 2 ) · · · (q − x n )
(x 1 x 2 · · · x n )n−1 ≥ .
(q − 1)n
and
x 1 + x 2 + · · · + x n n
(q + x 1 )(q + x 2 ) · · · (q + x n ) ≤ q + = (q + 1)n .
n
The equality occurs for a1 = a2 = · · · = an = 1.
2n − 1
Remark. For p = , we get the following inequality
(n − 1)2
n
1
≤ 1,
i=1 (n − 1)2 + (2n − 1)ai
which holds for all positive numbers a1 , a2 , . . . , an (n ≥ 3) satisfying a1 a2 · · · an = 1.
where
1
a= (a1 + a2 + · · · + an ).
n
(Darij Grinberg and Vasile Cîrtoaje, 2006)
Solution. Let
1
bi = aj, i = 1, 2, . . . , n.
n − 1 j =i
By Jensen’s inequality, we have
ai − ai+1 ai + (n − 1)bi+1 1 1
f a+ =f ≤ f (ai ) + 1 − f (bi+1 ),
n n n n
hence
n ai − ai+1
n n
n f a+ ≤ f (ai ) + (n − 1) f (bi ).
i=1
n i=1 i=1
a1 a2 · · · an = 1,
then
a1 a2 an
2(a1n + a2n + · · · + ann ) + n(n − 2) ≥ n + + ··· + .
a2 a3 a1
x 1 + x 2 + · · · + x n = 0.
f (x) = e x , x ∈ ,
we get
x1 −x2 x 2 −x 3 x n −x 1
2e x 1 + 2e x 2 + · · · + 2e x n + n(n − 2) ≥ n e n + e n + · · · + e n .
x 1 = n ln a1 , x 2 = n ln a2 , . . . , x n = n ln an ,
a1 = a2 = · · · = an = 1.
a1 + a2 + · · · + an = n,
then a1 − a2 a2 − a3 a n − a1 2
1+ 1+ ··· 1 + ≥ (a1 a2 · · · an ) n .
n n n
f (x) = − ln x, x > 0,
we get
. a1 − a2 a2 − a3 an − a1 /
n ln 1 + + ln 1 + + · · · + ln 1 + ≥
n n n
≥ 2(ln a1 + ln a2 + · · · + ln an ),
which is equivalent to the desired inequality. The equality occurs if and only if
a1 = a2 = · · · = an = 1.
On Popoviciu’s Inequality 521
a1 + a2 + · · · + an = n,
then
2 1 1 1 1 1 1
+ + ··· + +n−2≥ + + ··· + .
n a1 a 2 an a1 − a2 a2 − a3 an − a1
1+ 1+ 1+
n n n
1
f (x) = , x > 0.
x
The equality occurs if and only if a1 = a2 = · · · = an = 1.
P 5.17. Let f be a convex function on (0, ∞). If a1 , a2 , . . . , an are positive real num-
bers, then
1 1 1
f a1 + + f a2 + + · · · + f an + ≥
a2 a3 a1
1 1 1
≥ f a1 + + f a2 + + · · · + f an + .
a1 a2 an
(Vasile C., 2009)
and ) *
1 1 1
a1 + ≥ max a1 + , a2 + ,
a2 a1 a2
the inequality for n = 2 follows from Lemma below (which is a consequence of
Karamata’s inequality). To prove the original inequality, consider that
an+1 ≤ min{a1 , a2 , . . . , an },
522 Vasile Cîrtoaje
and use the induction method. Based on the induction hypothesis, we only need
to show that
1 1 1 1
f an + + f an+1 + ≥ f an + + f an+1 + .
an+1 a1 a1 an+1
This inequality follows also by Lemma below, since
1 1 1 1
an + + an+1 + = an + + an+1 +
an+1 a1 a1 an+1
and ) *
1 1 1
an + ≥ max an + , an+1 + .
an+1 a1 an+1
The equality occurs for a1 = a2 = · · · = an .
Lemma. Let f be a convex function on a real interval . If a, b, c, d ∈ such that
a + b = c + d, a ≥ max{c, d},
then
f (a) + f (b) ≥ f (c) + f (d).
Proof. Without loss of generality, assume that c ≥ d; then,
a ≥ c ≥ d ≥ b.
If a = c, then b = d, and the inequality is an equality. Consider now that a > c,
when
a > c ≥ d > b.
First proof. The desired inequality follows by adding the following Jensen’s inequal-
ities
(c − b) f (a) + (a − c) f (b) ≥ (a − b) f (c),
(a − c) f (a) + (c − b) f (b) ≥ (a − b) f (d).
Second Proof. Since c, d ∈ (b, a), there are p, q ∈ (0, 1) such that
c = pa + (1 − p)b, d = qa + (1 − q)b.
From a + b = c + d, we get
a + b = (p + q)a + (2 − p − q)b,
(a − b)(p + q − 1) = 0,
p + q = 1.
Using Jensen’s inequalities
f (c) ≤ p f (a) + (1 − p) f (b),
f (d) ≤ q f (a) + (1 − q) f (b),
we get
f (c) + f (d) ≤ (p + q) f (a) + (2 − p − q) f (b) = f (a) + f (b).
Appendix A
Glosar
p1 + p2 + · · · + pn = 1.
523
524 Vasile Cîrtoaje
5. BERNOULLI’S INEQUALITY
−1 ≤ a1 , a2 , . . . , an ≤ 0,
then
(1 + a1 )(1 + a2 ) · · · (1 + an ) ≥ 1 + a1 + a2 + · · · + an .
6. SCHUR’S INEQUALITY
For any nonnegative real numbers a, b, c and any positive number k, the inequality
holds
a k (a − b)(a − c) + b k (b − c)(b − a) + c k (c − a)(c − b) ≥ 0,
with equality for a = b = c, and for a = 0 and b = c (or any cyclic permutation).
For k = 1, we get the third degree Schur’s inequality, which can be rewritten as
follows
a3 + b3 + c 3 + 3a bc ≥ a b(a + b) + bc(b + c) + ca(c + a),
(a + b + c)3 + 9a bc ≥ 4(a + b + c)(a b + bc + ca),
9a bc
a2 + b2 + c 2 + ≥ 2(a b + bc + ca),
a+b+c
Glosar 525
a4 + b4 + c 4 − a2 b2 − b2 c 2 − c 2 a2 ≥ (a b + bc + ca)(a2 + b2 + c 2 − a b − bc − ca),
(b − c)2 (b + c − a)2 + (c − a)2 (c + a − b)2 + (a − b)2 (a + b − c)2 ≥ 0,
6a bc p ≥ (p2 − q)(4q − p2 ), p = a + b + c, q = a b + bc + ca.
A generalization of the fourth degree Schur’s inequality, which holds for any
real numbers a, b, c and any real number m, is the following (Vasile Cirtoaje, 2004)
(a − mb)(a − mc)(a − b)(a − c) ≥ 0,
with equality for a = b = c, and also for a/m = b = c (or any cyclic permutation).
This inequality is equivalent to
a4 + m(m + 2) a2 b2 + (1 − m2 )a bc a ≥ (m + 1) a b(a2 + b2 ),
(b − c)2 (b + c − a − ma)2 ≥ 0.
7. CAUCHY-SCHWARZ INEQUALITY
If a1 , a2 , . . . , an and b1 , b2 , . . . , bn are real numbers, then
8. HÖLDER’S INEQUALITY
If x i j (i = 1, 2, · · · , m; j = 1, 2, · · · n) are nonnegative real numbers, then
n m
(
m n 0(
m
≥
m
xi j xi j .
i=1 j=1 j=1 i=1
526 Vasile Cîrtoaje
9. CHEBYSHEV’S INEQUALITY
Let a1 ≥ a2 ≥ · · · ≥ an be real numbers.
a) If b1 ≥ b2 ≥ · · · bn , then
n
n
n
n ai bi ≥ ai bi ;
i=1 i=1 i=1
b) If b1 ≤ b2 ≤ · · · ≤ bn , then
n
n
n
n ai bi ≤ ai bi .
i=1 i=1 i=1
and
and
b1 + · · · + bi ≥ c1 + · · · + ci , i = 1, 2, · · · , n.
If a1 ≥ a2 ≥ · · · ≥ an ≥ 0, then
a1 b1 + a2 b2 + · · · + an bn ≥ a1 c1 + a2 c2 + · · · + an cn .
Notice that all these inequalities follow immediately from the identity
i
n
n
i
ai (bi − ci ) = (ai − ai+1 ) bj − cj , an+1 = 0.
i=1 i=1 j=1 j=1
Glosar 527
p = a + b + c, q = a b + bc + ca, r = a bc,
s = p2 − 3q = a2 + b2 + c 2 − a b − bc − ca.
From the identity
it follows that
−2p3 + 9pq − 2(p2 − 3q) p2 − 3q −2p3 + 9pq + 2(p2 − 3q) p2 − 3q
≤r≤ ,
27 27
which is equivalent to
A = (a1 , a2 , . . . , an ), ai ∈ ,
B = (b1 , b2 , . . . , bn ), bi ∈ ,
then
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ f (b1 ) + f (b2 ) + · · · + f (bn ).
We say that a sequence A = (a1 , a2 , . . . , an ) with a1 ≥ a2 ≥ · · · ≥ an majorizes a
sequence B = (b1 , b2 , . . . , bn ) with b1 ≥ b2 ≥ · · · ≥ bn , and write it as
A B,
528 Vasile Cîrtoaje
if
a1 ≥ b1 ,
a1 + a2 ≥ b1 + b2 ,
·····················
a1 + a2 + · · · + an−1 ≥ b1 + b2 + · · · + bn−1 ,
a1 + a2 + · · · + an = b1 + b2 + · · · + bn .
f (αx + β y) ≤ α f (x) + β f ( y)
Right Half Convex Function Theorem (Vasile Cîrtoaje, 2004). Let f be a real
function defined on an interval and convex on ≥s , where s ∈ int(). The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
holds for all a1 , a2 , . . . , an ∈ satisfying
a1 + a2 + · · · + an = ns
if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ such that x ≤ s ≤ y and x + (n − 1) y = ns.
Left Half Convex Function Theorem (Vasile Cîrtoaje, 2004). Let f be a real function
defined on an interval and convex on ≤s , where s ∈ int(). The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
Glosar 529
a1 + a2 + · · · + an = ns
if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ such that x ≥ s ≥ y and x + (n − 1) y = ns.
Right Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real
function defined on an interval and convex on [s, s0 ], where s, s0 ∈ , s < s0 . In
addition, f is decreasing on ≤s0 and f (u) ≥ f (s0 ) for u ∈ . The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
holds for all a1 , a2 , . . . , an ∈ satisfying
a1 + a2 + · · · + an = ns
if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ such that x ≤ s ≤ y and x + (n − 1) y = ns.
Left Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real
function defined on an interval and convex on [s0 , s], where s0 , s ∈ , s0 < s. In
addition, f is increasing on ≥s0 and f (u) ≥ f (s0 ) for u ∈ . The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
holds for all a1 , a2 , . . . , an ∈ satisfying
a1 + a2 + · · · + an = ns
if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ such that x ≥ s ≥ y and x + (n − 1) y = ns.
Equal Variables Theorem for Nonnegative Variables (Vasile Cirtoaje, 2005). Let
a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let
0 ≤ x1 ≤ x2 ≤ · · · ≤ x n
such that
x 1 x 2 · · · x n = a1 a2 · · · an .
Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )
is maximum for
x 1 = x 2 = · · · = x n−1 ≤ x n ,
and is minimum for
0 < x1 ≤ x2 = x3 = · · · = x n
or
0 = x 1 = · · · = x j ≤ x j+1 ≤ x j+2 = · · · = x n , j ∈ {1, 2, . . . , n − 1}.
Equal Variables Theorem for Real Variables (Vasile Cirtoaje, 2010). Let a1 , a2 , . . . , an
(n ≥ 3) be fixed real numbers, and let
0 ≤ x1 ≤ x2 ≤ · · · ≤ x n
such that
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,
Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )
Best Upper Bound of Jensen’s Difference Theorem (Vasile Cirtoaje, 1990). Let
p1 , p2 , . . . , pn (n ≥ 3) be fixed positive real numbers, and let f be a convex function
on = [a, b]. If a1 , a2 , . . . , an ∈ , then Jensen’s difference
p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) p1 a1 + p2 a2 + · · · + pn an
−f
p1 + p2 + · · · + pn p1 + p2 + · · · + pn
F (a1 , a2 , . . . , an ) : → , ∈ n ,
be a symmetric continuous function satisfying
a + a a1 + an
1 n
F (a1 , a2 , . . . , an−1 , an ) ≥ F , a2 , . . . , an−1 ,
2 2
for all a1 , a2 , . . . , an ∈ such that a1 ≤ a2 ≤ · · · ≤ an or a1 ≥ a2 ≥ · · · ≥ an . Then,
for all a1 , a2 , . . . , an ∈ , the following inequality holds:
a1 + a2 + · · · + an
F (a1 , a2 , . . . , an ) ≥ F (a, a, . . . , a), a= .
n
F (a1 , a2 , . . . , an ) : → , ∈ n ,
be a symmetric continuous function satisfying
a + a a1 + an−1
1 n−1
F (a1 , a2 , . . . , an−2 , an−1 , an ) ≥ F , a2 , . . . , an−2 , , an
2 2
for all a1 , a2 , . . . , an ∈ such that a1 ≤ a2 ≤ · · · ≤ an (or a1 ≥ a2 ≥ · · · ≥ an ). Then,
for all a1 , a2 , . . . , an ∈ such that a1 ≤ a2 ≤ · · · ≤ an (or a1 ≥ a2 ≥ · · · ≥ an ), the
following inequality holds:
a1 + a2 + · · · + an−1
F (a1 , a2 , . . . , an−1 , an ) ≥ F (t, t, . . . , t, an ), t= .
n−1
Arithmetic Compensation Theorem (Vasile Cîrtoaje, 2005). Let s > 0 and let F be
a symmetric continuous function on the compact set in n
S = {(a1 , a2 , . . . , an ) : a1 + a2 + · · · + an = s, ai ≥ 0, i = 1, 2, . . . , n}.
If
F (a1 , a2 , a3 , . . . , an ) ≥
a + a a + a
1 2 1 2
≥ min F , , a3 , . . . , an , F (0, a1 + a2 , a3 , . . . , an ) (*)
2 2
for all (a1 , a2 , . . . , an ) ∈ S, then
s s
F (a1 , . . . , an−k , an−k+1 , . . . , an ) ≥ min F 0, . . . , 0, , . . . ,
1≤k≤n k k
for all (a1 , a2 , . . . , an ) ∈ S.
532 Vasile Cîrtoaje
Arithmetic Compensation Corollary (Vasile Cîrtoaje, 2005). Let s > 0 and let F be
a symmetric continuous function on the compact set in n
S = {(a1 , a2 , . . . , an ) : a1 + a2 + · · · + an = s, ai ≥ 0, i = 1, 2, . . . , n}.
If
F (a1 , a2 , a3 , . . . , an ) ≥ F (0, a1 + a2 , a3 , . . . , an )
for all (a1 , a2 , . . . , an ) ∈ S satisfying
a + a a + a
1 2 1 2
F (a1 , a2 , a3 , . . . , an ) < F , , a3 , . . . , a n , a1 = a2 ,
2 2
then F (a1 , a2 , . . . , an ) is minimum when n − k of the variables a1 , a2 , . . . , an are zero
s
and the other k variables are equal to , where k ∈ {1, 2, . . . , n}.
k
a + b + c = p, a b + bc + ca = q,
where p and q are fixed real numbers satisfying p2 ≥ 3q, then the product r = a bc is
minimum for a = b, and maximum for b = c.
a + b + c = p, a b + bc + ca = q,
where p and q are fixed real numbers satisfying p2 ≥ 3q, then the product r = a bc is
minimum for a = b or c = 0, and maximum for b = c.
a + b + c = p, a bc = r,
p = a + b + c, q = a b + bc + ca, r = a bc.
f6 (a, 1, 1) ≥ 0
Theorem 3 (Vasile Cîrtoaje, 2008). Let f6 (a, b, c) be a sixth degree symmetric homo-
geneous polynomial having the highest coefficient A ≤ 0. The inequality f6 (a, b, c) ≥
0 holds for all nonnegative real numbers a, b, c if and only if f6 (a, 1, 1) ≥ 0 and
f6 (0, b, c) ≥ 0 for all nonnegative real numbers a, b, c.
∗∗∗∗∗
ξ ∈ (−∞, 0) ∪ (3, ∞)
∗∗∗∗∗
holds for all nonnegative real numbers a, b, c if and only if it holds for b = c = 1
and for a = 0 (see Theorem 3);
3) treating successively the cases p2 < 4q and p2 ≥ 4q.
Index
A
AC-Corollary: 6, 20-27, 30-32, 34-38, 42-44, 46, 47, 51, 53-55, 58, 87-90, 92-96.
adding: 235, 305, 508, 522.
AM-Corollary: 3, 4, 18-22, 35, 36, 43-45, 49-52, 77, 79, 81, 83-86.
AM1-Corollary: 4, 62, 63, 66-71, 74, 75.
AM-GM inequality: 23, 24, 26, 28, 30, 31, 50, 96, 112, 183, 411, 430, 508, 510,
517, 518.
AM-Theorem: 3, 4, 17.
any permutation: 32, 33, 39, 40, 55, 59, 61, 62, 65, 73, 74, 87, 90, 92.
assume: 17, 41, 43, 46, 47, 50, 77, 79, 80, 82, 84-86, 88, 89, 91, 93, 96, 112,
126, 132, 134, 135, 137, 138, 141, 143, 145, 149, 150, 152, 154, 157, 159, 161,
163, 165, 180, 183, 292, 392, 499, 512, 516, 517, 521, 522, 530.
assumption: 2, 51, 53, 502.
C
cancellation: 186-188, 359, 361.
Cauchy-Schwarz inequality: 240, 270, 327, 354, 511, 513, 525.
Chebyshev’s inequality: 526.
claim: 38, 53, 251.
concave: 198, 528.
consequence: 3, 34, 37, 58, 74, 168, 170-172, 254, 341, 500, 521.
consequently: 49, 514.
contradiction: 512, 513, 516, 517.
convex: 499-502, 508-511, 513, 514, 516, 518-522, 527, 528-530, 534.
D
decreasing: 49, 67, 255, 256, 276, 278, 279, 297, 298, 311, 500, 516, 526, 527,
529.
537
538 Vasile Cîrtoaje
decreasing: 49, 67, 255, 256, 276, 278, 279, 297, 298, 311, 500, 516, 526, 527,
529.
degree eight: 197, 198, 356, 368, 482, 483, 486-488, 490, 492, 494, 497.
degree seven: 368, 369.
degree six: 185-188, 192, 193, 359, 361, 364, 365, 367, 482, 483, 486, 487, 533,
534.
G
generalization: 25, 29 32, 139, 175, 217, 221, 222, 233, 241, 248, 253, 262, 263,
279, 288, 292, 294, 314, 317, 327, 329, 334, 335, 379, 381-384, 386, 388, 435,
438, 441, 447, 452, 464, 468, 471, 479, 481, 482, 485, 487, 491, 493, 496, 498,
501, 517, 525.
H
highest coefficient: 186-188, 192, 193, 198, 213-220, 221, 223-234, 236, 237,
239, 242, 246, 250, 254, 257-259, 261, 263-265, 267, 269-271, 273, 275, 277,
280, 282-285, 287, 291, 294, 295, 301, 303, 307, 313, 316, 317, 319, 321-323,
326-328, 331-333, 335-342, 344, 359-362, 364, 365, 367, 368, 379-385, 387, 389,
390, 393-396, 398, 419, 420, 431, 434, 436, 439, 443, 445, 450, 454, 455, 457,
459, 461, 463, 465, 467, 469-474, 477-479, 482, 483, 486, 487, 533-534.
highest polynomial: 197, 198, 346-348, 350, 351, 353, 354, 356, 368, 369, 469,
470, 472, 474, 477-479, 482, 483, 486-488, 490, 492, 494, 497.
homogeneity (homogeneous): 17, 39, 41, 43, 45, 48, 50, 52, 69, 75, 116-118,
127, 129, 131, 132, 134, 135, 137-139, 141, 143, 146, 149, 151, 152, 154, 157,
159, 161, 163, 166, 171, 173, 175, 177, 180, 183, 185, 187-189, 192, 193, 197,
198, 340, 359-361, 364, 365, 367-369, 488, 507, 515, 533, 534.
hypothesis: 1, 6, 21, 23, 24, 26, 28, 30, 31, 64, 72, 113, 138, 193, 500, 510, 513,
522.
I
identity: 48, 194, 195, 227, 231, 234, 235, 253, 256, 300, 305, 306, 312, 366,
398, 401,485, 526, 527.
increasing: 48, 49, 255, 256, 278, 298, 524.
induction: 5, 500, 522.
interval: 2, 499, 500, 501, 505, 519, 522, 527-529, 534.
J
Jensen’s inequality: 500, 501, 518, 519, 522.
K
Karamata’s inequality: 500, 501, 521, 527.
known: 3, 174, 176, 178, 472, 507.
Index 539
L
left inequality: 109, 281, 315, 338, 381, 516.
lemma: 2, 3,6, 100, 101, 185, 187, 198, 359, 360, 521, 522.
loss: 77, 79, 522.
M
maximum: 98-100, 109, 185, 198, 359, 527, 530, 532.
minimum: 5-7, 38, 99, 100, 109-111, 185, 198, 359, 502, 527, 530, 532.
more general: 56, 119, 123, 127, 128, 315, 350, 351.
multiplying: 72, 174, 403, 419, 421, 484, 518.
N
nontrivial: 48, 88, 132, 134, 135, 137, 138, 141, 143, 388, 486, 489, 491, 493,
515.
notation: 437.
O
obvious: 6, 19, 21, 42, 44-46, 78, 80, 81, 83, 85-88, 111, 115, 122, 125, 129,
177, 179, 431.
P
Popoviciu’s inequality: 499-501, 507, 509-511, 513, 514, 516, 518, 519, 535.
power mean: 5244.
proportional to: 118-121, 123-125, 127, 130, 133, 135, 136, 138, 142, 144, 146,
150, 152, 153, 155, 158, 160, 162, 164, 167, 179, 181, 260, 263.
R
real function: 528, 529.
remark: 25, 28, 29, 32, 39, 41, 42, 111, 115, 122, 125, 129, 139, 141, 143, 145,
147, 148, 150, 153, 156, 158, 161, 163, 165, 168, 169, 171, 172, 175, 179, 187,
194, 195, 197-199, 227, 231, 234, 236, 238, 243, 260, 263, 275, 277, 178, 316,
320, 329, 355-368, 507, 511, 515, 517, 519.
replacing: 1, 2, 132, 134, 135, 137, 138, 141, 143, 145, 149, 150, 152, 154, 157,
159, 161, 163, 165, 180, 183, 315, 368, 398, 513.
right inequality: 516, 528, 529.
root: 118-125, 127, 130, 133, 135, 136, 138, 139, 142, 144, 146, 148, 150,
152-155, 158, 160, 162, 164, 167, 175, 179, 181, 182, 184, 260, 263.
S
same manner: 25, 29, 32, 517.
same highest: 187, 213-216, 218, 220, 221, 223-233, 236, 237, 239, 246, 258,
261, 264-267, 269-271, 273, 280, 282, 283, 285, 287, 291, 294, 301, 307, 313,
540 Vasile Cîrtoaje
316, 317, 319, 321-323, 331-333, 335, 336, 339, 346, 347, 350, 351, 353, 354,
356, 379, 380, 382, 385, 387, 389, 390, 394, 396, 398, 431, 439, 443, 450, 457,
459, 461, 463, 465, 467, 469, 470, 472, 474, 477, 488, 490, 492, 494, 497.
Schur’s inequality: 39, 524, 525.
sharper: 111, 115, 122, 125, 129, 156, 168-171.
squaring: 76,167.
substitution (substituting): 18, 36, 48, 59, 119, 120, 123, 124, 126, 128, 146,
149, 151, 155, 157, 160, 162, 164, 166, 183, 191, 235, 253, 256, 300, 304, 306,
312, 384, 386, 388, 398, 401, 434, 440, 446, 448, 451, 454, 456, 458, 459, 462,
464, 466, 468, 473, 507, 508, 512, 516, 517, 520.
summing: 111, 115, 122, 125, 129, 499.
U
unchanged: 132, 134, 135, 137, 138, 141, 143, 145, 149, 150, 152, 154, 157,
159, 161, 163, 165, 180, 183, 398.
V
vicinity: 196, 222, 257, 281, 314, 349, 354
Bibliography
[1] Andreescu T., Cîrtoaje V., Dospinescu G., Lascu M., Old and New Inequalities,
GIL Publishing House, 2004.
[2] Bin X., Boreico I., Can V.Q.B., Bulj A., Lascu M., Opympiad Inequalities, GIL
Publishing House, 2015.
[3] Bin X., Boreico I., Can V.Q.B., Cîrtoaje V., Lascu M., An Introduction to Inequal-
ities, GIL Publishing House, 2015.
[4] Can V.Q.B., Pohoaţă C., Old and New Inequalities, GIL Publishing House, 2008.
[5] Can V.Q.B., Anh T.Q., Su Dung Phuong Phap Cauchy-Schwarz De Chung Minh
Bat Dang Thuc, Nha Xuat Ban Dai Hoc Su Pham, 2010.
[8] Cîrtoaje V., Algebraic Inequalities-Old and New Methods, GIL Publishing House,
2006.
[10] Cîrtoaje V., The Equal Variable Method, Journal of Inequalities In Pure and
Applied Mathematics, Volume 8, Issue 1, 2007.
[11] Cîrtoaje V., On Jensen Type Inequalities with Ordered Variables, Journal of In-
equalities In Pure and Applied Mathematics, Volume 9, Issue 1, 2008.
[12] Cîrtoaje V., The Proof of Three Open Inequalities, Crux Mathematicorum, Vol-
ume 34, Issue 4, 2008.
[13] Cîrtoaje V., Can V.Q.B., Anh T.Q., Inequalities with Beautiful Solutions, GIL
Publishing House, 2009.
541
542 Vasile Cîrtoaje
[14] Cîrtoaje V., On the Cyclic Homogeneous Polynomial Inequalities of Degree Four,
Journal of Inequalities in Pure and Applied Mathematics, Volume 10, Issue 3,
2009.
[15] Cîrtoaje V., The Best Upper Bound for Jensen’s Inequality, Australian Journal of
Mathematical Analysis and Aplications, Volume 7, Issue 2, Art. 22, 2011.
[16] Cîrtoaje V., Baiesu A., An Extension of Jensen’s Discrete Inequality to Half Convex
Functions, Journal of Inequalities and Applications, Volume 2011.
[18] Cîrtoaje V., Can V.Q.B., On Some Cyclic Homogeneous Polynomial Inequalities
of Degree Four in Real Variables, International Journal of Pure and Applied
Mathematics, Volume 80, No. 3, 2012.
[19] Cîrtoaje V., The Best Lower Bound for Jensen’s Inequality with three fixed ordered
variables Journal, Banach Journal of Mathematical Analysis, Volume 7, Issue
1, 2013.
[20] Cîrtoaje V., An Extension of Jensen’s Discrete Inequality to Partially Convex Func-
tions, Journal of Inequalities and Applications, Volume 2013:54.
[21] Cîrtoaje V., A Strong Method for Symmetric Homogeneous Polynomial Inequali-
ties of Degree Six in Nonnegative Real Variables, British Journal of Mathemati-
cal and Computers Science, 4(5), 2014.
[22] Cîrtoaje V., On the Equal Variables Method Applied to Real Variables, Creative
Mathematics and Informatics, no. 2, 2015.
[23] Cîrtoaje V., Three Extensions of HCF and PCF Theorems, Advances in Inequali-
ties and Applications, no. 2, 2016.
[24] Cîrtoaje V., Extensions of Jensen’s Discrete Inequality with Ordered Variables
to Half and Partially Convex Functions, Journal of Inequalities and Special
Functions, Volume 8, Issue 3, 2017.
[26] Cîrtoaje V., Mathematical Inequalities - Volume 2, Symmetric Rational and Non-
rational Inequalities, Lambert Academic Publishing, 2018.
[27] Cîrtoaje V., Mathematical Inequalities - Volume 3, Cyclic and Noncyclic Inequal-
ities, Lambert Academic Publishing, 2018.
[30] Hung P.K., Secrets in Inequalities, Volume 1: Basic Inequalities, GIL Publishing
House, 2007.
[31] Hung P.K., Secrets in Inequalities, Volume 2: Advanced Inequalities, GIL Pub-
lishing House, 2008.
[32] Littlewood G.H., Polya J.E., Inequalities, Cambridge University Press, 1967.
[33] Mitrinovic̆ D.S., Pecaric̆ J.E., Fink A.M., Classical and New Inequalities in Anal-
ysis, Kluwer, 1993.
View publication stats